Hugging Face's logo

Datasets:
hitoshura25
/
megavul

Tasks:
Text Generation
Modalities:
Text
Formats:
parquet
Languages:
code
Size:
100K - 1M
ArXiv:
Tags:
security
vulnerabilities
code-analysis
cve
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
megavul / README.md
hitoshura25's picture
hitoshura25
Upload README.md with huggingface_hub
dbca97f verified
preview code
|
raw
history blame contribute delete
2.91 kB
metadata 
license: gpl-3.0
task_categories:
  - text-generation
  - code-generation
language:
  - code
size_categories:
  - 10K<n<100K
pretty_name: MegaVul (CVEfixes Format)
tags:
  - security
  - vulnerabilities
  - code-analysis
  - cve

MegaVul Dataset (CVEfixes-Compatible Format)

Dataset Description

This is a processed version of the MegaVul dataset converted to match the CVEfixes schema format for unified vulnerability analysis and model training.

Source Kaggle Dataset: marcdamie/megavul-a-cc-java-vulnerability-dataset Original Project: Icyrockton/MegaVul

Dataset Summary

MegaVul is a large, high-quality, extensible, continuously updated C/C++/Java function-level vulnerability dataset.

This processed version provides:

  • 671,797 vulnerability records
  • 8,476 unique CVEs
  • 9,284 unique repositories
  • CVEfixes-compatible schema for combined training

Languages

Language Count Percentage
C 538269 80.1%
C++ 133528 19.9%

Dataset Structure

See the project's megavul_dataset_loader.py for full schema details.

Source Data

Original Dataset

This dataset is derived from the MegaVul project.

Data Collection

This dataset was created by:

  1. Downloading the version of the dataset from Kaggle.
  2. Converting it to the CVEfixes commit-level schema.
  3. Generating synthetic commit hashes and other metadata.

Licensing & Attribution

License

This dataset is licensed under GPL-3.0 to match the original MegaVul license.

Citation

If you use this dataset, please cite the original MegaVul paper:

@inproceedings{megavul2024,
  title={MegaVul: A C/C++ Vulnerability Dataset with Comprehensive Code Representations},
  author={Chen Ni and Zhuo Xu and Yun Zhang and Yang Liu and others},
  booktitle={Proceedings of the 21st International Conference on Mining Software Repositories},
  year={2024},
  publisher={ACM}
}

And acknowledge this processed version:

@misc{hitoshura25_megavul,
  title={MegaVul Dataset in CVEfixes Format},
  author={Your Name},
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/hitoshura25/megavul}},
  note={Processed from the Kaggle dataset by Marc Damie, based on the original MegaVul project.}
}

Usage 

from datasets import load_dataset

dataset = load_dataset("hitoshura25/megavul")