Pytorch implemenation of the SAFE neural network.

SAFE can be used to produce dense representations (i.e., embeddings) for arbitrary binary functions. It works for both the X86 and ARM architectures.

See our paper on arXiv: https://arxiv.org/abs/1811.05296

If you use this code, please cite:

@inproceedings{massarelli2018safe,
  title={SAFE: Self-Attentive Function Embeddings for Binary Similarity},
  author={Massarelli, Luca and Di Luna, Giuseppe Antonio and Petroni, Fabio and Querzoni, Leonardo and Baldoni, Roberto},
  booktitle={Proceedings of 16th Conference on Detection of Intrusions and Malware & Vulnerability Assessment (DIMVA)},
  year={2019}
}

(optional) It might be a good idea to use a separate conda environment. It can be created by running:

conda create -n safe37 -y python=3.7 && conda activate safe37
pip install -r requirements.txt

Download the model weights from http://dl.fbaipublicfiles.com/SAFEtorch/model.tar.gz

wget http://dl.fbaipublicfiles.com/SAFEtorch/model.tar.gz
tar -xzvf model.tar.gz
rm model.tar.gz

Please refer to this notebook test.ipynb

Or try out this script test.py to get all the function embeddings of the input binary.

python test.py <binary_path>

• SAFE implementation in tensorflow (https://github.com/gadiluna/SAFE)
• YARASAFE: Automatic Binary Function Similarity Checks with Yara (https://github.com/lucamassarelli/yarasafe)

SAFEtorch is licensed under the MIT license. The text of the license can be found here.