ARXIV:2604.04783 · ENCRYPTED AI INFERENCE · SUBMITTED 07 APR · 20:13 UTC · FRESHNESS UNKNOWN

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference

Guoci Chen · Xiurui Pan · Qiao Li · Bo Mao · Congming Gao · Chengying Huan · +2 at arXiv

Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

Blocked on Code›Score4.0Evidence unverified

Opportunity summary

Pain Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

Evidence 0 refs | 0 sources | 0% coverage

Blocker Evidence unverified

Open Build Read PDF Signal Canvas Track

PROBLEM

Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers. Fully Homomorphic Encryption (FHE) allows computation on encrypted data, but current FHE methods struggle with efficient and precise nonlinear function evaluation.

METHOD

Full abstract

Deploying large language models (LLMs) as cloud services raises privacy concerns as inference may leak sensitive data. Fully Homomorphic Encryption (FHE) allows computation on encrypted data, but current FHE methods struggle with efficient and precise nonlinear function evaluation. Specifically, CKKS-based approaches require high-degree polynomial approximations, which are costly when target precision increases. Alternatively, TFHE's Programmable Bootstrapping (PBS) outperforms CKKS by offering exact lookup-table evaluation. But it lacks high-precision implementations of LLM nonlinear layers and underutilizes GPU resources. We propose \emph{TIGER}, the first GPU-accelerated framework for high-precision TFHE-based nonlinear LLM layer evaluation. TIGER offers: (1) GPU-optimized WoP-PBS method combined with numerical algorithms to surpass native lookup-table precision limits on nonlinear functions; (2) high-precision and efficient implementations of key nonlinear layers, enabling practical encrypted inference; (3) batch-driven design exploiting inter-input parallelism to boost GPU efficiency. TIGER achieves 7.17$\times$, 16.68$\times$, and 17.05$\times$ speedups over a CPU baseline for GELU, Softmax, and LayerNorm, respectively.

RESULT

ScienceToStartup currently rates this 4.0/10 on the public viability pass. TIGER achieves 7.17$\times$, 16.68$\times$, and 17.05$\times$ speedups over a CPU baseline for GELU, Softmax, and LayerNorm, respectively.

WHY NOW

Encrypted AI Inference moved forward this cycle; last verified April 2026. Public score 4.0/10.

Continue into Read for claims, analysis, references, and neighboring papers.

Opportunity summary

Score4.0

PainAccelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

Evidence0 refs | 0 sources | 0% coverage

Blockerno shell-level blocker reported

Analysis summary

Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

Competitive landscape

Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

Segment

Encrypted AI Inference

Adoption evidence

No public code link in the paper record yet

Commercial read

4.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

{ "contract_version": "paper-r2", "paper_id": "545578d5-752d-4e8e-b75d-3d89bc51125d", "arxiv_id": "2604.04783", "canonical_route": "/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "active_tab": "synced from current hash by the drawer client", "selected_artifact": "gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "endpoints": { "paper_pack": "/api/v1/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference/paper-pack", "build_passport": "/api/v1/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference/build-passport", "mcp_resource": "sciencetostartup://surfaces/paper-workspace" } }

{ "surface": "paper", "mode": "paper", "query": "GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference", "normalized_query": "2604.04783", "route": "/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "paper_ref": "gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "topic_slug": null, "benchmark_ref": null, "dataset_ref": null }

{ "@context": "https://schema.org", "@graph": [ { "@type": "WebPage", "@id": "https://sciencetostartup.com/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference#webpage", "url": "https://sciencetostartup.com/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "name": "GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference", "description": "Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.", "isPartOf": { "@id": "https://sciencetostartup.com/#website" } }, { "@type": "ScholarlyArticle", "@id": "https://sciencetostartup.com/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference#scholarlyArticle", "headline": "GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference", "description": "Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.", "url": "https://sciencetostartup.com/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference", "sameAs": "https://arxiv.org/abs/2604.04783", "identifier": { "@type": "PropertyValue", "propertyID": "arXiv", "value": "2604.04783" }, "isAccessibleForFree": true, "isPartOf": { "@id": "https://sciencetostartup.com/#website" }, "datePublished": "2026-04-06T15:54:35.000Z", "author": [ { "@type": "Person", "name": "Guoci Chen" }, { "@type": "Person", "name": "Xiurui Pan" }, { "@type": "Person", "name": "Qiao Li" }, { "@type": "Person", "name": "Bo Mao" }, { "@type": "Person", "name": "Congming Gao" }, { "@type": "Person", "name": "Chengying Huan" }, { "@type": "Person", "name": "Mingzhe Zhang" }, { "@type": "Person", "name": "Jie Zhang" } ], "additionalProperty": [ { "@type": "PropertyValue", "propertyID": "viabilityScore", "value": 4 }, { "@type": "PropertyValue", "propertyID": "researchDomain", "value": "Encrypted AI Inference" } ] }, { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://sciencetostartup.com" }, { "@type": "ListItem", "position": 2, "name": "Encrypted AI Inference", "item": "https://sciencetostartup.com/topics" }, { "@type": "ListItem", "position": 3, "name": "GPU Acceleration of TFHE-Based High-Precision Nonlinear Laye", "item": "https://sciencetostartup.com/paper/gpu-acceleration-of-tfhe-based-high-precision-nonlinear-layers-for-encrypted-llm-inference" } ] } ] }

Competitive landscape

Accelerating encrypted LLM inference with a GPU-optimized framework for high-precision nonlinear layers.

Segment

Encrypted AI Inference

Adoption evidence

No public code link in the paper record yet

Commercial read

4.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference

GPU Acceleration of TFHE-Based High-Precision Nonlinear Layers for Encrypted LLM Inference

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline