albertan017/LLM4Decompile

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

6986 stars · 548 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

LLM4Decompile is an open-source series of large language models (1.3B to 33B) trained to decompile binary code back into readable, executable high-level source code, outperforming traditional tools like Ghidra. The repository includes model weights, training scripts, and the HumanEval-Decompile and decompile-bench datasets for training and evaluation.

Use cases

• decompile binary executables back into C source code
• recover readable source from stripped or optimized binaries
• improve on Ghidra's unreadable decompiler output with LLMs
• train a custom decompilation model on binary-source function pairs
• benchmark LLM decompilation with HumanEval-Decompile
• analyze malware or legacy binaries without source code

When to choose

• you need ML-based decompilation with more readable and re-executable output than Ghidra
• you want open-source decompiler models you can fine-tune or self-host
• you are researching LLM applications in reverse engineering
• you need large binary-source paired datasets for training

When to avoid

• you need a production-grade, guaranteed-correct decompiler for security audits
• you lack GPU resources to run multi-billion-parameter models
• you need decompilation for languages or architectures outside the training data
• you want a lightweight traditional decompiler without ML overhead

Facets

library · maturity active

reverse-engineering machine-learning llm-inference llm-training parser reverse-engineering large-language-models security machine-learning python decompilation binary-analysis ghidra disassembly research-models huggingface linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="albertan017/LLM4Decompile")