Browse by hardware¶
Declared in extracted details¶
Values reported by the structured extraction (field hardware).
- NVIDIA A100 GPU (5): THOR: Secure transformer inference with homomorphic encryption; MOAI: Module-optimizing architecture for non-interactive secure transformer inference; TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU; ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs; Large language model inference acceleration: A comprehensive hardware perspective
- CPU (4): BumbleBee: Secure Two-party Inference Framework for Large Transformers; CKKS Bootstrapping in less than 10ms; Privft: Private and fast text classification with homomorphic encryption; Large language model inference acceleration: A comprehensive hardware perspective
- GPU (4): SoK: Private LLM Inference using Approximate Homomorphic Encryption; SIGMA: Secure GPT Inference with Function Secret Sharing; Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware; Large language model inference acceleration: A comprehensive hardware perspective
- NVIDIA A100 80GB GPU (3): STIP: Efficient and secure non-interactive transformer inference via compact packing; ENSI: Efficient non-interactive secure inference for large language models; Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference
- NVIDIA H200 GPU (2): ELLMo: Packing- and Depth-Aware Encrypted Transformer Inference; MOAI: Module-optimizing architecture for non-interactive secure transformer inference
- NVIDIA RTX 4090 GPU (2): FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs; Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures
- 1.5TB DDR5 (1): Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage
- 2-socket Intel Xeon Gold 6242 2.80GHz, 64 cores (1): Private and reliable neural network inference
- 2.4 GHz Intel Core2 6600, x86-64 (1): Fully Homomorphic Encryption with Relatively Small Key and Ciphertext Sizes
- 2.9 GHz CPU, single-threaded (1): A Full RNS Variant of Approximate Homomorphic Encryption
- 2x AMD EPYC 9334 CPU (1): Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage
- 2x Intel Xeon E5-2690 v4 (56 logical cores, 925 GB RAM) (1): Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption
- 2x Intel Xeon Platinum 8272L (104 logical cores, 2975 GB RAM) (1): Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption
- 2x NVIDIA H100 NVL (94GB HBM each) (1): Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage
- 3.7 GHz Intel Xeon, 4 cores, 16 GB RAM (1): CrypTFlow2: Practical 2-Party Secure Inference
- 32-core CPU (1): SHIP: A Shallow and Highly Parallelizable CKKS Bootstrapping Algorithm
- 3x Alibaba Cloud ecs.g7.8xlarge (Intel Xeon Platinum 8369B, 32 vCPU, 128 GB RAM) (1): PUMA: Secure Inference of LLaMA-7B in Five Minutes
- 3x Alibaba Cloud ecs.g7.xlarge instances (4 vCPU, 16 GB RAM); LAN 10.1 Gbps/0.1 ms, WAN 300 Mbps/40 ms (1): SecretFlow-SPU: A Performant and User-Friendly Framework for Privacy-Preserving Machine Learning
- 3x Tesla V100 servers, 10 GB/s bandwidth (1): SecFormer: Fast and Accurate Privacy-Preserving Inference for Transformer Models via SMPC
- 48-core server with 4x AMD Opteron 6172 and 64 GB RAM (1): Stream ciphers: A Practical Solution for Efficient Homomorphic-Ciphertext Compression
- 5 Gbps LAN, 1 ms RTT (1): PUMA: Secure Inference of LLaMA-7B in Five Minutes
- 5G network (60/100 Mbps) (1): PrivateLoRA for Efficient Privacy Preserving LLM
- 64-core machine (batch-latency sweep) (1): Machine Learning Training on Encrypted Data with TFHE
- 7th-generation Intel i7 CPU (2.8 GHz base frequency) (1): MP-SPDZ: A Versatile Framework for Multi-Party Computation
- 8-core CPU with 16 threads (1): Machine Learning Training on Encrypted Data with TFHE
- 8x NVIDIA GeForce RTX 4090 24GB (1): Encryption-friendly LLM architecture
- 8x NVIDIA P100 GPU (1): Attention is all you need
- 8x Samsung 990 PRO 2TB PCIe Gen5 SSDs (two RAID-0 arrays) (1): Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage
- A100 80GB GPU (1): PrivateLoRA for Efficient Privacy Preserving LLM
- Alibaba Cloud ecs.c7.2xlarge instances (2.70 GHz CPU, 16 GB RAM) (1): Cheetah: Lean and Fast Secure Two-Party Deep Neural Network Inference
- AMD EPYC 7542 CPU (32 threads) (1): STIP: Efficient and secure non-interactive transformer inference via compact packing
- AMD EPYC 7542 CPU @ 2.90 GHz, 32 threads, 128 GB RAM (1): Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity
- AMD EPYC 9V84 CPU (SEV-SNP) (1): Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study
- AMD MI100 GPU (1): Large language model inference acceleration: A comprehensive hardware perspective
- AMD Ryzen 7 2700X @ 3.70 GHz (single-threaded), 64 GB memory (1): Rubato : Noisy Ciphers for Approximate Homomorphic Encryption (Full Version)
- AMD Ryzen 9 5900X 12-core CPU, 64 GB RAM (1): Activate Me!: Designing Efficient Activation Functions for Privacy-Preserving Machine Learning with Fully Homomorphic Encryption
- AMD Ryzen 9 7950X3D CPU (1): HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0
- AMD Ryzen AI MAX+ 395 CPU, 128 GB RAM, single thread (1): FHE for SIMD Arithmetic Logic Units with Amortized O(1) Bootstrapping per Ciphertext
- AMD Ryzen CPU (32 cores, 3.60GHz), 256GB RAM (1): Ditto: Quantization-aware Secure Inference of Transformers upon MPC
- AMD Ryzen Threadripper 3990X 64-core CPU (1): Towards Case-Optimized Hybrid Homomorphic Encryption
- AMD Ryzen Threadripper 7970X (32 cores), 157 GB RAM (1): Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference
- AMD Ryzen Threadripper 7970X CPU (32 cores, 160 GB RAM), single thread (1): Low-Depth Bootstrapping for Matrix-Native FHE
- Apple M1 Max (MacBook Pro 2021, 64 GB RAM) (1): Fully Homomorphic Encryption for Cyclotomic Prime Moduli
- Apple M3 Max CPU, 48 GB unified memory (1): Revisiting Shamir Secret Sharing for Threshold Fully Homomorphic Encryption
- Apple M4 Max, 128 GB RAM (single thread) (1): Efficient Homomorphic Integer Computer from CKKS
- Apple M4 Max, 128 GB RAM, single-threaded (1): Bootstrapping GBFV with CKKS
- Apple M4 Pro (single thread, 24 GB memory) (1): Stay in Your Lane: Fast Arithmetic over Large Finite Rings with CKKS
- Apple M4 Pro CPU (single thread, 32 GiB RAM) (1): Leveraging Discrete CKKS to Bootstrap in High Precision
- ASIC (1): Large language model inference acceleration: A comprehensive hardware perspective
- ASIC (proposed accelerator design) (1): ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse
- ASIC, 12 nm GlobalFoundries, 1 GHz, 150 mm2 (simulated, not yet fabricated) (1): BASALISC: Programmable Hardware Accelerator for BGV Fully Homomorphic Encryption
- AWS c4.2xlarge (8 vCPUs, 15 GiB) (1): TenSEAL: A Library for Encrypted Tensor Operations Using Homomorphic Encryption
- AWS c4.4xlarge (16 vCPUs, 30 GiB) (1): TenSEAL: A Library for Encrypted Tensor Operations Using Homomorphic Encryption
- AWS c5.2xlarge (Intel Xeon 8000 series 3.0GHz, 16GB RAM), client and server in different regions (1): Delphi: A Cryptographic Inference Service for Neural Networks
- AWS c5n.metal instance (single thread) (1): One server for the price of two: Simple and fast single-server private information retrieval
- AWS c6i.2xlarge instances (8 vCPUs, 16 GiB RAM) per party (1): Ajax: Fast Threshold Fully Homomorphic Encryption without Noise Flooding
- AWS EC2 c6i.16xlarge (64 vCPUs, 128 GB RAM), 32 threads (1): BOLT: Privacy-Preserving, Accurate and Efficient Inference for Transformers
- AWS single-region bandwidth setting (1): FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System
- BTS ASIC (373.6 mm^2, up to 163.2 W) (1): BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption
- Cloud TPU (1): BERT: Pre-training of deep bidirectional transformers for language understanding
- commodity laptop (1): Private svm inference on encrypted data
- CPU baseline (1): BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption
- CPU with AVX (1): Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware
- CPU, up to 32 processors (1): PaCo: Bootstrapping for CKKS via Partial CoeffToSlot
- dual AMD EPYC 7763 CPUs (128 cores), 1 TB RAM (1): CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption
- dual Intel Xeon Platinum 8280 CPU (112 cores), 172 GB memory (1): Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network
- F1 accelerator (cycle-accurate simulation and RTL synthesis, 16 compute clusters, 2x HBM2) (1): F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption
- flagship smartphone chipset (1): PrivateLoRA for Efficient Privacy Preserving LLM
- four distinct GPU platforms (Table IV) (1): FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs
- FPGA (1): Large language model inference acceleration: A comprehensive hardware perspective
- general-purpose CPU (1): Private pathological assessment via machine learning and homomorphic encryption
- GF12LP ASIC multi-project wafer (150 mm2 target) (1): TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation
- Google Cloud TPU (1): HellaSwag: Can a machine really finish your sentence?
- Google Colab (1): Federated Learning: An approach with Hybrid Homomorphic Encryption
- GPU (CUDA, model not stated in excerpt) (1): zkLLM: Zero Knowledge Proofs for Large Language Models
- GPU (HEaaN GPU-accelerated library, model not specified) (1): DaCapo: Automatic Bootstrapping Management for Efficient Fully Homomorphic Encryption
- GPU (model not specified in excerpt) (1): Secure transformer inference made non-interactive
- GPU (model not stated in excerpt) (1): Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs
- GPU (model unspecified in excerpt) (1): EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption
- GPU (unspecified in excerpt) (1): EncFormer: Secure and Efficient Transformer Inference over Encrypted Data
- GPU (via FIDESlib CKKS backend; model not named in excerpt) (1): HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation
- H100 GPU (1): Theory, analysis, and best practices for sigmoid selfattention
- Intel Core i5 2.9 GHz CPU (1): Homomorphic Encryption for Arithmetic of Approximate Numbers
- Intel Core i5-10210U CPU (1): ModHE: Modular Homomorphic Encryption Using Module Lattices: Potentials and Limitations
- Intel Core i7 2.8 GHz laptop CPU, 16 GB RAM, single thread (1): Improved Bootstrapping for Approximate Homomorphic Encryption
- Intel Core i7-10700 CPU (16 GB RAM) (1): Large-Plaintext Functional Bootstrapping in FHE with Small Bootstrapping Keys
- Intel Core i7-3520M CPU (2.9 GHz) (1): Improved Security for a Ring-Based Fully Homomorphic Encryption Scheme
- Intel Core i7-4910MQ 2.90GHz laptop, single thread (1): Multi-Key Homomorphic Encryption from TFHE
- Intel Core i7-7700HQ CPU at 2.80 GHz (1): Fully Homomorphic Encryption on the Ring of Gaussian Periods
- Intel Core i9-14900K CPU (32 cores) (1): ENSI: Efficient non-interactive secure inference for large language models
- Intel Core i9-14900KF CPU (3.20 GHz), 128 GB RAM, single-threaded (1): High-Precision Functional Bootstrapping for CKKS from Fourier Extension
- Intel i5-6600k 3.5 GHz, 32 GB RAM, single-threaded (1): Efficient Bootstrapping for Approximate Homomorphic Encryption with Non-Sparse Keys
- Intel i5-6600K 3.50 GHz, 32 GB RAM, single-threaded (Windows 10) (1): Bootstrapping for Approximate Homomorphic Encryption with Negligible Failure-Probability by Using Sparse-Secret Encapsulation
- Intel i7-7700 8-core 3.60 GHz CPU with 32 GB RAM (1): FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System
- Intel SGX (1): Privacypreserving llm inference in practice: A comparative survey of techniques, trade-offs, and deployability
- Intel SGX enclave (1): Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware
- Intel Skylake 18-core 1.6GHz CPU (1): CrypTen: Secure Multi-Party Computation Meets Machine Learning
- Intel TDX (1): Privacypreserving llm inference in practice: A comparative survey of techniques, trade-offs, and deployability
- Intel Xeon E-2176M @ 4.00 GHz, single thread, 32 GB (ThinkPad P1) (1): Efficient Multi-Key Homomorphic Encryption with Packed Ciphertexts with Application to Oblivious Neural Network Inference
- Intel Xeon E5-1630 v4 3.70GHz, 8 cores (CPU only) (1): Privacy-Preserving Graph-Based Machine Learning with Fully Homomorphic Encryption for Collaborative Anti-Money Laundering
- Intel Xeon Gold 6130 CPU (1): InferDPT: Privacy-Preserving Inference for Closed-box Large Language Model
- Intel Xeon Gold 6234 CPU (8 cores) (1): Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs
- Intel Xeon Gold 6240R CPU (2.40 GHz, 48 threads, 384 GB RAM) (1): ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation
- Intel Xeon Gold 6242 2.8 GHz, 503 GiB RAM, single-threaded (1): Bootstrapping Bits with CKKS
- Intel Xeon Gold 6242 CPU 2.80GHz (single thread, AVX512) (1): Plaintext-Ciphertext Matrix Multiplication and FHE Bootstrapping: Fast and Fused
- Intel Xeon Gold 6258R CPU (1 TB RAM) (1): Large-Plaintext Functional Bootstrapping in FHE with Small Bootstrapping Keys
- Intel Xeon Platinum 8173M (16 vCPUs, 128 GB RAM) (1): MPC-Minimized Secure LLM Inference
- Intel Xeon Platinum 8268 2.90 GHz CPU, 192 GB RAM, single thread (1): Accelerating HE Operations from Key Decomposition Technique
- Intel Xeon Platinum 8268 CPU @ 2.90GHz, 376 GB RAM (1): Multi-key Fully Homomorphic Encryption with Non-Interactive Setup in the Plain Model
- Intel Xeon Platinum 8369B 64-core 2.50 GHz CPU, 128 GB RAM (1): XBOOT: Free-XOR Gates for CKKS with Applications to Transciphering
- Intel Xeon Platinum 8473C CPU (52 cores, 104 threads) (1): THEMIS: A Co-Designed System for Encrypted Transformer Inference
- Intel Xeon Platinum 8480+ CPU (56 cores) (1): MOAI: Module-optimizing architecture for non-interactive secure transformer inference
- Intel Xeon Platinum 8558 CPU (TDX) (1): Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study
- Intel Xeon Platinum 8558P CPU (192 threads) (1): Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache
- Intel Xeon Silver 4114 CPU @ 2.20 GHz (single thread) (1): Efficient Homomorphic Evaluation on Large Intervals
- Intel Xeon Silver 4208 CPU @ 2.10GHz, 30GB RAM (1): Non-interactive Fully Encrypted Machine Learning Protocol for Inference
- LAN 377 MBps / 0.3 ms (1): CrypTFlow2: Practical 2-Party Secure Inference
- LAN 5Gbps / 0.4ms RTT (1): Ditto: Quantization-aware Secure Inference of Transformers upon MPC
- MacBook Air, Apple M1 (1): Revisiting Module Lattice-based Homomorphic Encryption and Application to Secure-MPC
- MacBook Pro, Intel Core i9, 4 cores at 2.3 GHz (1): Secure outsourced matrix computation and application to neural networks
- multi-core CPU platform with 104 threads (model not specified) (1): DOReN: Toward Efficient Deep Convolutional Neural Networks with Fully Homomorphic Encryption
- NVIDIA A100 80 GB GPU (1): ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation
- NVIDIA A100 GPU (single) (1): Powerformer: Efficient and high-accuracy privacy-preserving language model with homomorphic encryption
- NVIDIA A100 PCIe (80 GB) (1): Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference
- NVIDIA A100-80GB GPU (1): Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache
- NVIDIA A100-SXM4 80GB GPU with 24-core AMD EPYC 7543 (1): Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption
- NVIDIA B200 GPUs (1-4) (1): Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B)
- NVIDIA GeForce RTX 4090 GPU (1): HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0
- NVIDIA GeForce RTX 5090 (32 GB) (1): Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference
- NVIDIA GPU (1): Privft: Private and fast text classification with homomorphic encryption
- NVIDIA GPU (CUDA) (1): Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library
- NVIDIA GPUs (multi-GPU) (1): A scalable multi-GPU framework for encrypted large-model inference
- NVIDIA H100 80 GB GPU (1): An Open-Source End-to-End FHE Implementation for Privacy-Preserving Llama 3 8B Inference
- NVIDIA H100 Confidential Computing (1): Privacypreserving llm inference in practice: A comparative survey of techniques, trade-offs, and deployability
- NVIDIA H100 GPU (1): Large language model inference acceleration: A comprehensive hardware perspective
- NVIDIA H100 NVL GPU (1): Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study
- NVIDIA H200 NVL GPU (1): Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study
- NVIDIA P100 GPU (1): CrypTen: Secure Multi-Party Computation Meets Machine Learning
- NVIDIA RTX 4090 24GB GPU (1): Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference
- NVIDIA RTX 6000 Ada 48GB GPU (1): Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity
- NVIDIA RTX A6000 GPU (1): InferDPT: Privacy-Preserving Inference for Closed-box Large Language Model
- NVIDIA RTX PRO 6000 GPU (1): CKKS Bootstrapping in less than 10ms
- NVIDIA RTX PRO 6000 GPUs (1-8) (1): Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B)
- NVIDIA Tesla T4 GPU (1): FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System
- NVIDIA Tesla V100 GPU (1): Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs
- NVIDIA Tesla V100 GPUs (ML experiments) (1): Delphi: A Cryptographic Inference Service for Neural Networks
- NVIDIA V100 GPU (16 GB) (1): MPC-Minimized Secure LLM Inference
- Palladium emulation system (1): TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation
- PIM/NDP (1): Large language model inference acceleration: A comprehensive hardware perspective
- PYNQ-Z2 FPGA (1): HHEML: Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge
- RTX 6000 GPU (1): Fast and Accurate Homomorphic Softmax Evaluation
- single 64-bit Intel core at 3 GHz (1): FHEW: Bootstrapping Homomorphic Encryption in Less Than a Second
- single GPU (model not stated in excerpt) (1): TRICYCLE: Private transformer inference with tricyclic encodings
- single mid-range personal computer (1): Practical Privacy-Preserving Machine Learning using Fully Homomorphic Encryption
- single-core CPU (unspecified) (1): TFHE: Fast Fully Homomorphic Encryption over the Torus
- single-thread CPU (1): Fast and Accurate Homomorphic Softmax Evaluation
- Snapdragon 8 Gen 3 (1): Large language model inference acceleration: A comprehensive hardware perspective
- Tesla V100 SXM2 16GB GPU (1): [Skyformer: Remodel selfattention with gaussian kernel and nystr](../papers/skyformer-remodel-selfattention-with-gaussian-kernel-and-nystr.md)
- V100 GPU (1): fairseq: A fast, extensible toolkit for sequence modeling
- WAN 40 MBps / 80 ms (1): CrypTFlow2: Practical 2-Party Secure Inference
- WAN 400Mbps / 40ms RTT (simulated with tc) (1): Ditto: Quantization-aware Secure Inference of Transformers upon MPC
- Xeon E3-1240v5 4-core 3.5 GHz CPU baseline (1): F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption
- Xilinx Alveo U280 FPGA (1): ModHE: Modular Homomorphic Encryption Using Module Lattices: Potentials and Limitations
- Xilinx U280 FPGA (1): Large language model inference acceleration: A comprehensive hardware perspective
Mentions in full text¶
Papers ordered by how often the term occurs (count in brackets). Only papers with a local PDF are counted.
GPU (137 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (205), A scalable multi-GPU framework for encrypted large-model inference (154), Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage (137), Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs (112), HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0 (103), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (103), SIGMA: Secure GPT Inference with Function Secret Sharing (97), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (81), Theory, analysis, and best practices for sigmoid selfattention (80), SoK: Private LLM Inference using Approximate Homomorphic Encryption (77), Privft: Private and fast text classification with homomorphic encryption (57), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (52), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (51), Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs (50), SoK: Private transformer-based model inference (49), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (48), Towards efficient privacy-preserving machine learning: A systematic review from protocol, model, and system perspectives (45), STIP: Efficient and secure non-interactive transformer inference via compact packing (43), CrypTen: Secure Multi-Party Computation Meets Machine Learning (40), Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference (39), EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (39), SoK: Wildest dreams: Reproducible research in privacy-preserving neural network training (32), Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study (30), PrivateLoRA for Efficient Privacy Preserving LLM (27), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (23), ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs (21), MOAI: Module-optimizing architecture for non-interactive secure transformer inference (21), A Survey of Deep Learning Architectures for Privacy-Preserving Machine Learning With Fully Homomorphic Encryption (20), Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption (20), TRICYCLE: Private transformer inference with tricyclic encodings (20), CKKS Bootstrapping in less than 10ms (19), ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour (19), Privacypreserving llm inference in practice: A comparative survey of techniques, trade-offs, and deployability (18), Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity (18), fairseq: A fast, extensible toolkit for sequence modeling (17), Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware (17), Efficiency Optimization Techniques in Privacy-Preserving Federated Learning With Homomorphic Encryption: A Brief Survey (16), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (15), Secure transformer inference made non-interactive (15), Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities (14), ENSI: Efficient non-interactive secure inference for large language models (13), ARION: Attention-optimized transformer inference on encrypted data (13), EncFormer: Secure and Efficient Transformer Inference over Encrypted Data (12), Llama: Open and efficient foundation language models (11), ELLMo: Packing- and Depth-Aware Encrypted Transformer Inference (11), Approximate homomorphic encryption based privacy-preserving machine learning: a survey (10), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (10), An Open-Source End-to-End FHE Implementation for Privacy-Preserving Llama 3 8B Inference (10), Encryption-friendly LLM architecture (10), SHAFT: Secure, Handy, Accurate, and Fast Transformer Inference (9), Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache (9), Attention is all you need (8), Private pathological assessment via machine learning and homomorphic encryption (7), F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption (7), DaCapo: Automatic Bootstrapping Management for Efficient Fully Homomorphic Encryption (7), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (7), Delphi: A Cryptographic Inference Service for Neural Networks (7), Fast and Accurate Homomorphic Softmax Evaluation (7), HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation (7), Error Propagation-Aware Scale Design for Efficient Homomorphic Encryption-Based LLM Inference (7) …
Intel Xeon (79 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (9), SoK: Wildest dreams: Reproducible research in privacy-preserving neural network training (7), SWIFT: Shallow and SIMD-Aware CKKS Functional Bootstrapping for Low-Latency (3), Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs (3), Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs (3), SoK: Private transformer-based model inference (2), Collaborative privacy-preserving analysis of oncological data using multiparty homomorphic encryption (2), Relaxed Functional Bootstrapping: A New Perspective on BGV/BFV Bootstrapping (2), Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network (2), Evaluation of Privacy-Preserving Support Vector Machine (SVM) Learning Using Homomorphic Encryption (2), Evaluation of Privacy-aware Support Vector Machine (SVM) Learning using Homomorphic Encryption (2), Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption (2), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (2), SHAFT: Secure, Handy, Accurate, and Fast Transformer Inference (2), TFHE Gets Real: An Efficient and Flexible Homomorphic Floating-Point Arithmetic (2), Efficient Homomorphic Evaluation on Large Intervals (2), Plaintext-Ciphertext Matrix Multiplication and FHE Bootstrapping: Fast and Fused (2), Bootstrapping Bits with CKKS (2), HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation (2), Practical secure inference algorithm for a fine-tuned large language model based on fully homomorphic encryption (2), MOAI: Module-optimizing architecture for non-interactive secure transformer inference (2), EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (2), ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs (1), Private and reliable neural network inference (1), SPIRAL: Fast, high-rate single-server PIR via FHE composition (1), Privft: Private and fast text classification with homomorphic encryption (1), ABY3: A Mixed Protocol Framework for Machine Learning (1), Rasta: A cipher with low ANDdepth and few ANDs per bit (1), Pasta: A Case for Hybrid Homomorphic Encryption (1), Multi-key Fully Homomorphic Encryption with Non-Interactive Setup in the Plain Model (1), Fully Homomorphic SIMD Operations (1), High-Precision Homomorphic ALU over Arbitrary Moduli with O (1) Bootstrapping (1), Large-Plaintext Functional Bootstrapping in FHE with Small Bootstrapping Keys (1), LFHE: Fully Homomorphic Encryption with Bootstrapping Key Size Less than a Megabyte (1), Efficient FHEW Bootstrapping with Small Evaluation Keys, and Applications to Threshold Homomorphic Encryption (1), XBOOT: Free-XOR Gates for CKKS with Applications to Transciphering (1), DOReN: Toward Efficient Deep Convolutional Neural Networks with Fully Homomorphic Encryption (1), Secure human action recognition by encrypted neural network inference (1), Non-interactive Fully Encrypted Machine Learning Protocol for Inference (1), Privacy-Preserving Graph-Based Machine Learning with Fully Homomorphic Encryption for Collaborative Anti-Money Laundering (1), Federated Learning: An approach with Hybrid Homomorphic Encryption (1), Touch of Privacy: A Homomorphic Encryption-Powered Deep Learning Framework for Fingerprint Authentication (1), BASALISC: Programmable Hardware Accelerator for BGV Fully Homomorphic Encryption (1), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (1), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (1), F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption (1), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (1), TenSEAL: A Library for Encrypted Tensor Operations Using Homomorphic Encryption (1), EVA: An Encrypted Vector Arithmetic Language and Compiler for Efficient Homomorphic Computation (1), Orion: A Fully Homomorphic Encryption Framework for Deep Learning (1), HeLayers: A Tile Tensors Framework for Large Neural Networks on Encrypted Data (1), OpenFHE: Open-Source Fully Homomorphic Encryption Library (1), InferDPT: Privacy-Preserving Inference for Closed-box Large Language Model (1), SecretFlow-SPU: A Performant and User-Friendly Framework for Privacy-Preserving Machine Learning (1), Centaur: Bridging the Impossible Trinity of Privacy, Efficiency, and Performance in Privacy-Preserving Transformer Inference (1), MPC-Minimized Secure LLM Inference (1), CipherGPT: Secure Two-Party GPT Inference (1), PUMA: Secure Inference of LLaMA-7B in Five Minutes (1), Iron: Private Inference on Transformers (1), CrypTFlow2: Practical 2-Party Secure Inference (1) …
CUDA (45 papers)¶
HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0 (16), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (9), zkLLM: Zero Knowledge Proofs for Large Language Models (8), A scalable multi-GPU framework for encrypted large-model inference (6), ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs (5), Towards efficient privacy-preserving machine learning: A systematic review from protocol, model, and system perspectives (5), Homomorphic Encryption for Machine Learning in Medicine and Bioinformatics (5), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (5), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (5), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (5), Privft: Private and fast text classification with homomorphic encryption (4), Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities (4), Large language model inference acceleration: A comprehensive hardware perspective (3), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (3), ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour (3), SoK: Wildest dreams: Reproducible research in privacy-preserving neural network training (2), CKKS Bootstrapping in less than 10ms (2), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (2), Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs (2), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (2), CrypTen: Secure Multi-Party Computation Meets Machine Learning (2), HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation (2), ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation (2), Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage (1), Theory, analysis, and best practices for sigmoid selfattention (1), Secure numerical simulations using fully homomorphic encryption (1), Cryptography-based privacy-preserving large language models: a lifecycle survey of frameworks, methods, and future directions (1), A Survey on Privacy-Preserving Machine Learning with Fully Homomorphic Encryption (1), Evaluation of Privacy-Preserving Support Vector Machine (SVM) Learning Using Homomorphic Encryption (1), Evaluation of Privacy-aware Support Vector Machine (SVM) Learning using Homomorphic Encryption (1), Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs (1), Efficiency Optimization Techniques in Privacy-Preserving Federated Learning With Homomorphic Encryption: A Brief Survey (1), Secure federated learning applied to medical imaging with fully homomorphic encryption (1), Touch of Privacy: A Homomorphic Encryption-Powered Deep Learning Framework for Fingerprint Authentication (1), Private pathological assessment via machine learning and homomorphic encryption (1), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (1), Orion: A Fully Homomorphic Encryption Framework for Deep Learning (1), Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study (1), BumbleBee: Secure Two-party Inference Framework for Large Transformers (1), SIGMA: Secure GPT Inference with Function Secret Sharing (1), EncFormer: Secure and Efficient Transformer Inference over Encrypted Data (1), STIP: Efficient and secure non-interactive transformer inference via compact packing (1), Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity (1), ARION: Attention-optimized transformer inference on encrypted data (1), SoK: Private LLM Inference using Approximate Homomorphic Encryption (1)
NVIDIA A100 (41 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (74), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (40), SoK: Private LLM Inference using Approximate Homomorphic Encryption (28), Theory, analysis, and best practices for sigmoid selfattention (22), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (21), PrivateLoRA for Efficient Privacy Preserving LLM (15), Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference (13), A scalable multi-GPU framework for encrypted large-model inference (11), Llama: Open and efficient foundation language models (10), ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs (7), HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation (7), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (6), EncFormer: Secure and Efficient Transformer Inference over Encrypted Data (6), Towards efficient privacy-preserving machine learning: A systematic review from protocol, model, and system perspectives (5), MOAI: Module-optimizing architecture for non-interactive secure transformer inference (5), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (4), Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption (3), Error Propagation-Aware Scale Design for Efficient Homomorphic Encryption-Based LLM Inference (3), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (2), F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption (2), Centaur: Bridging the Impossible Trinity of Privacy, Efficiency, and Performance in Privacy-Preserving Transformer Inference (2), Converting transformers to polynomial form for secure inference over homomorphic encryption (2), ENSI: Efficient non-interactive secure inference for large language models (2), Secure transformer inference made non-interactive (2), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (2), EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (2), High-Precision Exact FHE Made Simple, General, and Fast (1), Recent advances of privacy-preserving machine learning based on (Fully) Homomorphic Encryption (1), Touch of Privacy: A Homomorphic Encryption-Powered Deep Learning Framework for Fingerprint Authentication (1), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (1), zkLLM: Zero Knowledge Proofs for Large Language Models (1), CRAfT: Constant-Round Non-Polynomial Approximation and P2P Network for Secure Transformer Inference (1), SIGMA: Secure GPT Inference with Function Secret Sharing (1), Programmable Bootstrapping Enables Efficient Homomorphic Inference of Deep Neural Networks (1), ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation (1), An Open-Source End-to-End FHE Implementation for Privacy-Preserving Llama 3 8B Inference (1), STIP: Efficient and secure non-interactive transformer inference via compact packing (1), Powerformer: Efficient and high-accuracy privacy-preserving language model with homomorphic encryption (1), TRICYCLE: Private transformer inference with tricyclic encodings (1), Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache (1), THOR: Secure transformer inference with homomorphic encryption (1)
FPGA (29 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (97), HHEML: Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge (22), Efficiency Optimization Techniques in Privacy-Preserving Federated Learning With Homomorphic Encryption: A Brief Survey (16), ModHE: Modular Homomorphic Encryption Using Module Lattices: Potentials and Limitations (15), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (14), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (12), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (12), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (10), F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption (8), Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities (5), Recent advances of privacy-preserving machine learning based on (Fully) Homomorphic Encryption (4), Evaluation of Privacy-Preserving Support Vector Machine (SVM) Learning Using Homomorphic Encryption (3), BASALISC: Programmable Hardware Accelerator for BGV Fully Homomorphic Encryption (3), TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation (3), SoK: Private transformer-based model inference (2), Private pathological assessment via machine learning and homomorphic encryption (2), Accelerating HE Operations from Key Decomposition Technique (2), Secure numerical simulations using fully homomorphic encryption (1), A Survey of Deep Learning Architectures for Privacy-Preserving Machine Learning With Fully Homomorphic Encryption (1), Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network (1), Evaluation of Privacy-aware Support Vector Machine (SVM) Learning using Homomorphic Encryption (1), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (1), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (1), OpenFHE: Open-Source Fully Homomorphic Encryption Library (1), CipherGPT: Secure Two-Party GPT Inference (1), PEGASUS: Bridging Polynomial and Non-polynomial Evaluations in Homomorphic Encryption (1), SHIP: A Shallow and Highly Parallelizable CKKS Bootstrapping Algorithm (1), Secure transformer inference made non-interactive (1), ELLMo: Packing- and Depth-Aware Encrypted Transformer Inference (1)
ASIC (22 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (51), A scalable multi-GPU framework for encrypted large-model inference (29), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (20), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (13), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (9), BASALISC: Programmable Hardware Accelerator for BGV Fully Homomorphic Encryption (6), Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities (5), HHEML: Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge (5), TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation (5), F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption (5), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (3), Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware (3), SoK: Private transformer-based model inference (2), ModHE: Modular Homomorphic Encryption Using Module Lattices: Potentials and Limitations (2), OpenFHE: Open-Source Fully Homomorphic Encryption Library (2), Secure numerical simulations using fully homomorphic encryption (1), Approximate homomorphic encryption based privacy-preserving machine learning: a survey (1), Recent advances of privacy-preserving machine learning based on (Fully) Homomorphic Encryption (1), Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network (1), Private pathological assessment via machine learning and homomorphic encryption (1), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (1), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (1)
NVIDIA V100 (22 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (15), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (6), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (6), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (5), TensorFHE: Achieving Practical Computation on Encrypted Data Using GPGPU (4), Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs (4), Privft: Private and fast text classification with homomorphic encryption (3), Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs (3), fairseq: A fast, extensible toolkit for sequence modeling (2), SoK: Wildest dreams: Reproducible research in privacy-preserving neural network training (2), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (2), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (2), MPC-Minimized Secure LLM Inference (2), MPCFormer: Fast, Performant and Private Transformer Inference with MPC (2), [Skyformer: Remodel selfattention with gaussian kernel and nystr](../papers/skyformer-remodel-selfattention-with-gaussian-kernel-and-nystr.md) (1), Llama: Open and efficient foundation language models (1), Towards efficient privacy-preserving machine learning: A systematic review from protocol, model, and system perspectives (1), Secure human action recognition by encrypted neural network inference (1), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (1), zkLLM: Zero Knowledge Proofs for Large Language Models (1), SecFormer: Fast and Accurate Privacy-Preserving Inference for Transformer Models via SMPC (1), Delphi: A Cryptographic Inference Service for Neural Networks (1)
AMD EPYC (21 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (3), CKKS Bootstrapping in less than 10ms (2), Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage (1), High-Precision Exact FHE Made Simple, General, and Fast (1), The Power of Low Rank: Fast CKKS Functional Bootstrapping for High-Precision Lookup Tables (1), Building Privacy-Preserving AI system: AI Inference on Encrypted Data with Fully Homomorphic Encryption (1), Optimized Privacy-Preserving CNN Inference With Fully Homomorphic Encryption (1), Swift: Fast Secure Neural Network Inference With Fully Homomorphic Encryption (1), Practical and Private Hybrid ML Inference with Fully Homomorphic Encryption (1), Secure Federated Learning With Fully Homomorphic Encryption for IoT Communications (1), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (1), zkLLM: Zero Knowledge Proofs for Large Language Models (1), Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study (1), SHIP: A Shallow and Highly Parallelizable CKKS Bootstrapping Algorithm (1), General Functional Bootstrapping using CKKS (1), CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption (1), ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour (1), Converting transformers to polynomial form for secure inference over homomorphic encryption (1), STIP: Efficient and secure non-interactive transformer inference via compact packing (1), Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity (1), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (1)
NVIDIA RTX (19 papers)¶
Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (24), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (23), Large language model inference acceleration: A comprehensive hardware perspective (19), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (15), Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference (10), SoK: Private LLM Inference using Approximate Homomorphic Encryption (8), HellaSwag: Can a machine really finish your sentence? (2), Towards efficient privacy-preserving machine learning: A systematic review from protocol, model, and system perspectives (2), HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0 (2), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (2), ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour (2), Practical secure inference algorithm for a fine-tuned large language model based on fully homomorphic encryption (2), DaCapo: Automatic Bootstrapping Management for Efficient Fully Homomorphic Encryption (1), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (1), InferDPT: Privacy-Preserving Inference for Closed-box Large Language Model (1), An Open-Source End-to-End FHE Implementation for Privacy-Preserving Llama 3 8B Inference (1), Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity (1), Encryption-friendly LLM architecture (1), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (1)
Multi-GPU (17 papers)¶
A scalable multi-GPU framework for encrypted large-model inference (21), SoK: Private LLM Inference using Approximate Homomorphic Encryption (8), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (7), Privft: Private and fast text classification with homomorphic encryption (5), Terrazzo: Memory-Aware GPU Framework for Private Transformer Inference (4), Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage (3), fairseq: A fast, extensible toolkit for sequence modeling (3), HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0 (3), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (3), ELLMo: Packing- and Depth-Aware Encrypted Transformer Inference (2), Cryptography-based privacy-preserving large language models: a lifecycle survey of frameworks, methods, and future directions (1), Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library (1), PrivateLoRA for Efficient Privacy Preserving LLM (1), Zephyr: GPU-Efficient Homomorphic Encryption for Privacy-Preserving Transformer Inference (1), STIP: Efficient and secure non-interactive transformer inference via compact packing (1), Secure transformer inference made non-interactive (1), Encryption-friendly LLM architecture (1)
AVX-512 (12 papers)¶
OpenFHE: Open-Source Fully Homomorphic Encryption Library (8), SPIRAL: Fast, high-rate single-server PIR via FHE composition (5), Large language model inference acceleration: A comprehensive hardware perspective (2), Demystifying bootstrapping in fully homomorphic encryption (2), Efficiency Optimization Techniques in Privacy-Preserving Federated Learning With Homomorphic Encryption: A Brief Survey (2), FIDESlib: A Fully-Fledged Open-Source FHE Library for Efficient CKKS on GPUs (2), General Functional Bootstrapping using CKKS (2), FHE for SIMD Arithmetic Logic Units with Amortized O(1) Bootstrapping per Ciphertext (1), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (1), BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption (1), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (1), Plaintext-Ciphertext Matrix Multiplication and FHE Bootstrapping: Fast and Fused (1)
NVIDIA H100 (11 papers)¶
Theory, analysis, and best practices for sigmoid selfattention (22), Confidential Computing on NVIDIA Hopper GPUs: A Performance Benchmark Study (18), Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures (12), Cost-efficient llm training with lifetimeaware tensor offloading via gpudirect storage (10), Large language model inference acceleration: A comprehensive hardware perspective (9), A scalable multi-GPU framework for encrypted large-model inference (8), An Open-Source End-to-End FHE Implementation for Privacy-Preserving Llama 3 8B Inference (6), Privacypreserving llm inference in practice: A comparative survey of techniques, trade-offs, and deployability (4), SoK: Private LLM Inference using Approximate Homomorphic Encryption (4), Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B (v1: Scaling up Privacy-Preserving ML: A CKKS Implementation of Llama-2-7B) (3), ARK: Fully Homomorphic Encryption Accelerator with Runtime Data Generation and Inter-Operation Key Reuse (1)
Apple silicon (10 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (7), Stay in Your Lane: Fast Arithmetic over Large Finite Rings with CKKS (4), Enhancing privacy-preserving machine learning with self-learnable activation functions in fully homomorphic encryption (3), Revisiting Shamir Secret Sharing for Threshold Fully Homomorphic Encryption (1), Revisiting Module Lattice-based Homomorphic Encryption and Application to Secure-MPC (1), Bootstrapping GBFV with CKKS (1), Fully Homomorphic Encryption for Cyclotomic Prime Moduli (1), Efficient Homomorphic Integer Computer from CKKS (1), PaCo: Bootstrapping for CKKS via Partial CoeffToSlot (1), Leveraging Discrete CKKS to Bootstrap in High Precision (1)
ARM (10 papers)¶
A comprehensive survey on secure outsourced computation and its applications (9), Large language model inference acceleration: A comprehensive hardware perspective (3), Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware (3), HHEML: Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge (2), Rubato : Noisy Ciphers for Approximate Homomorphic Encryption (Full Version) (1), A Survey of Deep Learning Architectures for Privacy-Preserving Machine Learning With Fully Homomorphic Encryption (1), Building Privacy-Preserving AI system: AI Inference on Encrypted Data with Fully Homomorphic Encryption (1), TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation (1), Delphi: A Cryptographic Inference Service for Neural Networks (1), Grafting: Decoupled Scale Factors and Modulus in RNS-CKKS (1)
TPU (9 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (8), Split-and-Denoise: Protect Large Language Model Inference with Local Differential Privacy (8), BERT: Pre-training of deep bidirectional transformers for language understanding (7), HellaSwag: Can a machine really finish your sentence? (2), SecretFlow-SPU: A Performant and User-Friendly Framework for Privacy-Preserving Machine Learning (2), Practical Considerations of Fully Homomorphic Encryption in Privacy-Preserving Machine Learning (1), Touch of Privacy: A Homomorphic Encryption-Powered Deep Learning Framework for Fingerprint Authentication (1), CrypTen: Secure Multi-Party Computation Meets Machine Learning (1), A scalable multi-GPU framework for encrypted large-model inference (1)
NVIDIA Jetson (5 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (2), Approximate homomorphic encryption based privacy-preserving machine learning: a survey (2), Fully Homomorphically Encrypted Deep Learning as a Service (2), Touch of Privacy: A Homomorphic Encryption-Powered Deep Learning Framework for Fingerprint Authentication (1), PrivateLoRA for Efficient Privacy Preserving LLM (1)
NVIDIA T4 (5 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (3), Federated Learning: An approach with Hybrid Homomorphic Encryption (2), The HHE Land: Exploring the Landscape of Hybrid Homomorphic Encryption (1), Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network (1), FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System (1)
Raspberry Pi (4 papers)¶
Large language model inference acceleration: A comprehensive hardware perspective (3), Approximate homomorphic encryption based privacy-preserving machine learning: a survey (1), FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System (1), PrivateLoRA for Efficient Privacy Preserving LLM (1)