metadata
license: mit
This is the evaluation benchmark of LLM4Decompile project.
About
- Decompile-Bench-Eval includes manually crafted binaries from the well-established HumanEval and MBPP, alongside the compiled GitHub repositories released after 2025 to mitigate data leakage issues.
- C and C++ Support: These datasets include both C and C++ source code, whereas earlier models (LLM4Decompile-V1.5, V2) and the HumanEval-Decompile dataset were limited to C only.
Columns
It contains three splits, huameval, mbpp, and github2025. We also provide a json verison for the data. They contains the following columns:
{
"index":"index of the function",
"func_name":"demangled name for he function",
"func_dep":"function dependecies (includes, help functions), or the path to the source code",
"func":"source code",
"test":"unit tests for the function, empty for github data",
"opt":"optimization, O0, O1, O2, O3",
"language":"language, c or cpp",
"asm":"assembly",
"ida_asm":"assembly from ida pro",
"ida_pseudo":"decompiled results (pseudo code) from ida pro",
"ghidra_asm":"assembly from ghidra",
"ghidra_pseudo":"decompiled results (pseudo code) from ghidra"
}
Others
For more details, please check LLM4Decompile project.