README.md

metadata

license: gpl-3.0

CombinedVuln_LLVM Dataset

Dataset Summary

Combined_LLVM is a unified dataset for binary vulnerability detection that merges two sources:

• CompRealVul_LLVM, based on real-world C vulnerabilities
• Juliet_LLVM, based on the Juliet Test Suite

All functions are provided in LLVM Intermediate Representation (IR) format and labeled as vulnerable or non-vulnerable. This dataset is ideal for training and benchmarking machine learning models that aim to detect vulnerabilities in compiled representations of code, such as LLMs, GNNs, or binary-level classifiers.

Key Features

• ✅ LLVM IR format (field: llvm_ir_function) for architecture-independent representation
• ✅ Merged from two train datasets: Juliet (synthetic) and CompRealVul (real-world)
• ✅ Clear binary classification labels (label)
• ✅ Metadata included: source dataset, original file name, and function name

Dataset Structure

Each record in the dataset contains:

• dataset: The origin of the sample, either "Juliet" or "CompRealVul"
• file: The source filename of the C function
• fun_name: Name of the function
• llvm_ir_function: LLVM IR representation of the function
• label: "1" for vulnerable, "0" for non-vulnerable
• split: "train"

Example Entry

{
  "dataset": "Juliet",
  "file": "CWE121/s01.c",
  "fun_name": "CWE121_bad",
  "llvm_ir_function": "define dso_local void @CWE121_bad() { ... }",
  "label": "1",
  "split": "train"
}

Usage

from datasets import load_dataset

# Load the dataset
ds = load_dataset("compAgent/CombinedVuln_LLVM", split="train")

# View a sample
print(ds[0])

License

This dataset is released under the GPL-3.0.

Citation

@misc{combinedvuln_llvm,
  author       = {Compote},
  title        = {CombinedVuln_LLVM: A Unified Dataset of Vulnerable and Non-Vulnerable Functions in LLVM IR},
  howpublished = {\url{https://huggingface.co/datasets/compAgent/CombinedVuln_LLVM}},
  year         = {2025}
}