OPTango

OPTango: Multi-central Representation Learning against Innumerable Compiler Optimization for Binary Diffing

---

Description

This project consists of four subdirectories(model_weight, data, core, third) and one Python file(how_to_use.py).

-- model_weight: which contains the model and parameters of OPTango

-- data: which contains a small example data(preprocessed by IDA pro 7.5+) for validating OPTango's general functionality

-- core and third: which contain core functional code and script for preprocessing data

-- how_to_use.py: a script for validating OPTango's general functionality, such as extracting function features
and performing similarity matching between functions.

Environment Setup:

• Linux
• Python 3.8
• PyTorch 1.12.0
• CUDA 11.3
• IDA pro 7.5+ (only used for dataset processing)

where is data and model weight?

OPTBinary (Google Driver)

Optimization Settings of OPTBinary (Google Driver)

Model Weight (Google Driver)

1. download OPTBinary(bin.zip) and unzip as data/data-bsca/bin (not need if you just want to use OPTango)
2. download model_weight.zip and unzip as model_weight/

Getting Started:

a. Create a conda virtual environment and activate it.

conda create -n optango python=3.8 -y
conda activate optango

b. Install PyTorch and other packages.

conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt

c. Feature extracting and similarity matching with the OpenSSL demo data:

python how_to_use.py

d. Matching result The following diagram illustrates the cosine similarity between the query function "main" and nine index functions using OPTango:

f. refer to third/jTrans/datautils/README.md for preprocessing OPTBinary or your own data.

Normal issue

1. transformers version problem

Traceback (most recent call last):
  File "how_to_use.py", line 194, in <module>
    main()
  File "how_to_use.py", line 172, in main
    model = FullModel("cuda:0", with_gp=eval(args.with_gp))
  File "how_to_use.py", line 98, in __init__
    model.load_state_dict(checkpoint["bert"])
  File "/home/env_name/miniconda3/envs/optango2/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1604, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for OptRemoveBertModel:
        Unexpected key(s) in state_dict: "bert.embeddings.position_ids".

pip uninstall transformers
pip install transformers=4.28.0