Skip to content

Latest commit

 

History

History
47 lines (35 loc) · 1.7 KB

README.md

File metadata and controls

47 lines (35 loc) · 1.7 KB
Raw

Preface

This is the prototype implementation of our paper namely "History-Driven Fuzzing for Deep Learning Libraries" submitted to the FSE2024 conference. Please note that we have taken careful steps to protect the anonymity of the replication package.

Required Dependencies

The following are the dependencies required to run Orion:

inspect
pymongo
colorama
pandas

You can simply install all the required dependencies using the pip package manager.

Running Orion

Step 1

In order to run Orion, first you need to install the target DL library. For example, if you want to test Orion on PyTorch-1.13.1, you need to run the following command:

pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116

Please refer to the TensorFlow release history and PyTorch release history for further information.

Step 2

In the second step, you will need to download and extract the source in your desired directory.

Step 3

When you download the source code, you need to enter the following command in the terminal:

cd /root/fuzzing/

Then you have to enter the following command:

python run_fuzzer.py --database="database name" --library="target library" --release="target release" --tool="orion" --experiment_round=1

Example:

python run_fuzzer.py --database="orion-tf1" --library="tf" --release="2.11.0" --tool="orion" --experiment_round=1

Data

The reported vulnerabilities are available for TensorFlow and PyTorch.