LLM4Decompile: Decompiling Binary Code with Large Language Models

2024/03/08 by Hanzhuo Tan, Tan, Hanzhuo, Q. Luo +6 · 1 voice · 24 citations
Computer Science · Mathematics · #Arithmetic #Binary number #Code (set theory) #Computer science #Mathematics #Natural Language Processing Techniques #Programming language #Set (abstract data type) #cs.CL #cs.PL

paper · pdf · doi:10.48550/arxiv.2403.05286

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2024/03/08 · arxiv published 2024/03/08 · arxiv updated 2024/10/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Decompilation aims to convert binary code to high-level source code, but traditional tools like Ghidra often produce results that are difficult to read and execute. Motivated by the advancements in Large Language Models (LLMs), we propose LLM4Decompile, the first and largest open-source LLM series (1.3B to 33B) trained to decompile binary code. We optimize the LLM training process and introduce the LLM4Decompile-End models to decompile binary directly. The resulting models significantly outperform GPT-4o and Ghidra on the HumanEval and ExeBench benchmarks by over 100% in terms of re-executability rate. Additionally, we improve the standard refinement approach to fine-tune the LLM4Decompile-Ref models, enabling them to effectively refine the decompiled code from Ghidra and achieve a further 16.2% improvement over the LLM4Decompile-End. LLM4Decompile demonstrates the potential of LLMs to revolutionize binary code decompilation, delivering remarkable improvements in readability and executability while complementing conventional tools for optimal results. Our code, dataset, and models are released at https://github.com/albertan017/LLM4Decompile

Cited by

• When LLM Defenses Backfire: Characterizing Safety, Performance, and Cost Trade-offs
• Asm2SrcEval: Evaluating Large Language Models for Assembly-to-Source Code Translation
• Lares: LLM-driven Code Slice Semantic Search for Patch Presence Testing
• Exploring the Feasibility of End-to-End Large Language Model as a Compiler
• QiMeng-NeuComBack: Self-Evolving Translation from IR to Assembly Code
• Context-Guided Decompilation: A Step Towards Re-executability
• REx86: A Local Large Language Model for Assisting in x86 Assembly Reverse Engineering
• Binary Diff Summarization using Large Language Models
• Beyond Embeddings: Interpretable Feature Extraction for Binary Code Similarity
• SK2Decompile: LLM-based Two-Phase Binary Decompilation from Skeleton to Skin
• SoK: Potentials and Challenges of Large Language Models for Reverse Engineering
• SALT4Decompile: Inferring Source-level Abstract Logic Tree for LLM-Based Binary Decompilation
• D-LiFT: Improving LLM-based Decompiler Backend via Code Quality-driven Fine-tuning
• Decomposer: Learning to Decompile Symbolic Music to Programs
• AIRTBench: Measuring Autonomous AI Red Teaming Capabilities in Language Models
• Guaranteed Guess: A Language Modeling Approach for CISC-to-RISC Transpilation with Testing Guarantees
• Large Processor Chip Model
• Deconstructing Obfuscation: A four-dimensional framework for evaluating Large Language Models assembly code deobfuscation capabilities
• ReCopilot: Reverse Engineering Copilot in Binary Analysis
• Decompile-Bench: Million-Scale Binary-Source Function Pairs for Real-World Binary Decompilation
• DecompileBench: A Comprehensive Benchmark for Evaluating Decompilers in Real-World Scenarios
• ASSEMBLAGE-DEEPHISTORY: A Cross-Build Binary Dataset with Temporal Coverage
• Challenges and Future Directions in Agentic Reverse Engineering Systems
• The Digital Cybersecurity Expert: How Far Have We Come?

Discussions

• Sendo viável e factível, possivelmente no futuro grandes empresas poderão usar esse método para "roubar" os algoritmos de outros concorrentes! cc @qfoiluks.bsky.social @jpmartins.bsky.social [bsky, 2 points, 2 comments]

Related

• Decompile-Bench: Million-Scale Binary-Source Function Pairs for Real-World Binary Decompilation
• SK2Decompile: LLM-based Two-Phase Binary Decompilation from Skeleton to Skin
• SALT4Decompile: Inferring Source-level Abstract Logic Tree for LLM-Based Binary Decompilation
• D-LiFT: Improving LLM-based Decompiler Backend via Code Quality-driven Fine-tuning
• The CodeInverter Suite: Control-Flow and Data-Mapping Augmented Binary Decompilation with LLMs