Skip to content

LLM4Decompile

An open project for decompiling binary code with LLMs

Visit official site

What LLM4Decompile is

The project is described as an open-source effort to build and release language models dedicated to decompilation. The material states it compiles large numbers of C code samples into assembly code, then uses that data to fine-tune DeepSeek-Coder and build the Decompile-Eval benchmark.

What you can do with it

  • Compile C code samples into assembly code
  • Fine-tune the DeepSeek-Coder model on the compiled data
  • Build an evaluation benchmark from HumanEval questions and test samples
  • Evaluate decompilation through recompilability and executability

Application scenarios

  • Security researchers decompile suspected malware binaries for analysis
  • Developers recover source code from legacy binaries for maintenance and updates
  • Academic researchers explore decompilation techniques and program analysis methods

Pros & cons

✓ What we like

  • The project pairs a model with a dedicated benchmark
  • Recompilability and executability are named as evaluation goals

! What to watch out for

  • The described pipeline starts from compiled C code samples
  • The benchmark draws on HumanEval questions and test samples

FAQ

What is Decompile-Eval?

The material describes it as an evaluation benchmark built from compiled C code and HumanEval questions and test samples.

Which model is fine-tuned?

The material states the DeepSeek-Coder model is fine-tuned on the compiled data.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review