Skip to content
SAFE: Self-Attentive Function Embeddings for binary similarity
Python Ruby Perl Other
Branch: master
Clone or download
Luca Massarelli
Latest commit 1cd7b7a Sep 15, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
asm_embedding change license Jun 19, 2019
dataset_creation change license Jun 19, 2019
function_search fixed search engine evaluator Sep 15, 2019
neural_network change license Jun 19, 2019
Gemfile fixing site Nov 27, 2018
LICENSE Update Aug 5, 2019 no message Nov 27, 2018
_config.yml no message Nov 27, 2018 no message Nov 27, 2018
helloworld.c no message Nov 27, 2018
helloworld.o change license Jun 19, 2019
requirements.txt change license Jun 19, 2019

SAFE : Self Attentive Function Embedding


This software is the outcome of our accademic research. See our arXiv paper: arxiv

If you use this code, please cite our accademic paper as:

  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={To Appear in: Proceedings of 16th Conference on Detection of Intrusions and Malware & Vulnerability Assessment (DIMVA)},

What you need

You need radare2 installed in your system.


To create the embedding of a function:

git clone
pip install -r requirements
chmod +x
python -m data/safe.pb -i helloworld.o -a 100000F30

What to do with an embedding?

Once you have two embeddings embedding_x and embedding_y you can compute the similarity of the corresponding functions as:

from sklearn.metrics.pairwise import cosine_similarity

sim=cosine_similarity(embedding_x, embedding_y)

Data Needed

SAFE needs few information to work. Two are essentials, a model that tells safe how to convert assembly instructions in vectors (i2v model) and a model that tells safe how to convert an binary function into a vector. Both models can be downloaded by using the command


the downloader downloads the model and place them in the directory data. The directory tree after the download should be.

safe/-- githubcode

The safe.pb file contains the safe-model used to convert binary function to vectors. The i2v folder contains the i2v model.

Hardcore Details

This section contains details that are needed to replicate our experiments, if you are an user of safe you can skip it.


This is the freezed tensorflow trained model for AMD64 architecture. You can import it in your project using:

 import tensorflow as tf
 with tf.gfile.GFile("safe.pb", "rb") as f:
    graph_def = tf.GraphDef()

 with tf.Graph().as_default() as graph:
 sess = tf.Session(graph=graph)

see file: neural_network/


The i2v folder contains two files. A Matrix where each row is the embedding of an asm instruction. A json file that contains a dictonary mapping asm instructions into row numbers of the matrix above. see file: asm_embedding/

Train the model

If you want to train the model using our datasets you have to first use:

 python3 -td

This will download the datasets into data folder. Note that the datasets are compressed so you have to decompress them yourself. This data will be an sqlite databases. To start the train use neural_network/ The db can be selected by changing the parameter into If you want information on the dataset see our paper.

Create your own dataset

If you want to create your own dataset you can use the script ExperimentUtil into the folder dataset creation.

Create a functions knowledge base

If you want to use SAFE binary code search engine you can use the script ExperimentUtil to create the knowledge base. Then you can search through it using the script into function_search

Related Projects


In our code we use godown to download data from Google drive. We thank circulosmeos, the creator of godown.

We thank Davide Italiano for the useful discussions.

You can’t perform that action at this time.