TFHE-Coder: Evaluating LLM-agentic Fully Homomorphic Encryption Code Generation

2025-03-15

AI Summary

Generating correct, compilable TFHE code—particularly for logic gates and ReLU activation functions—is challenging due to high expert dependency, frequent errors, and low compilation success rates. Method: We propose the first LLM-driven integrated evaluation framework for TFHE compilation, incorporating RAG-enhanced few-shot prompting, a novel TFHE-specific encrypted code generation benchmark, tight integration with the tfhe-rs compiler, multi-source LLM inference (open- and closed-weight), and structural similarity quantification. Contribution/Results: Experiments show that off-the-shelf LLMs yield 72% error rates and only 31% compilation success; our framework reduces errors significantly, improves structural similarity by over 40%, and raises compilation success to 89%. This work provides the first systematic validation of LLMs’ feasibility and effectiveness in generating privacy-preserving, low-level homomorphic encryption code.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Search and Optimization: Heuristic Search

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements

Abstract

Fully Homomorphic Encryption over the torus (TFHE) enables computation on encrypted data without decryption, making it a cornerstone of secure and confidential computing. Despite its potential in privacy preserving machine learning, secure multi party computation, private blockchain transactions, and secure medical diagnostics, its adoption remains limited due to cryptographic complexity and usability challenges. While various TFHE libraries and compilers exist, practical code generation remains a hurdle. We propose a compiler integrated framework to evaluate LLM inference and agentic optimization for TFHE code generation, focusing on logic gates and ReLU activation. Our methodology assesses error rates, compilability, and structural similarity across open and closedsource LLMs. Results highlight significant limitations in off-the-shelf models, while agentic optimizations such as retrieval augmented generation (RAG) and few-shot prompting reduce errors and enhance code fidelity. This work establishes the first benchmark for TFHE code generation, demonstrating how LLMs, when augmented with domain-specific feedback, can bridge the expertise gap in FHE code generation.

Problem

• Evaluates LLM-based TFHE code generation for secure computing.
• Addresses cryptographic complexity and usability in FHE adoption.
• Establishes benchmarks for error reduction and code fidelity in FHE.

Innovation

• Compiler integrated framework for TFHE code generation
• Agentic optimizations like RAG and few-shot prompting
• Benchmark for TFHE code generation using LLMs