PagPassGPT

Codes for Paper: PagPassGPT: Pattern Guided Password Guessing via Generative Pretrained Transformer in DSN 2024.

1 Environment

conda create -n env_name python=3.8.10
conda activate env_name
pip install -r requirements.txt
# pip install numpy==1.24.2 huggingface-hub==0.13.4 fsspec==2022.11.0 torch==2.0.0 transformers==4.29.0 datasets==2.12.0 accelerate==0.17.1

2 Usage

2.1 Prepare datasets

1. You should have a dataset of passwords, like "RockYou" or other datasets. And you should make sure the dataset contains only passwords.

2. Run the script preprocess.sh to preprocess datasets.

sh ./scripts/preprocess.sh

Note: Here gives the "RockYou" dataset download link.

2.2 Train a PagPassGPT

Run the script train.sh to train.

sh ./scripts/train.sh

2.3 Generate passwords

Run the script generate.sh to generate.

sh ./scripts/generate.sh

Note: In this shell, you can choose to use D&C-GEN or not by changing just one line.

2.4 Evaluate passwords

sh ./scripts/evaluate.sh

Note: The evaluation mainly focus on Hit rate and Repeat rate.

3 Update Logs

• 2024.12.25: Fix some bugs.
• 2024.12.20: Add pip requirements.
• 2024.12.19: Update all codes.
  • Fix some bugs.
  • Provide more precise environmental requirements.
  • Provide new files for evaluation.
  • Make codes more user-friendly.
  • Update README.md.
• 2024.12.12: Update the paper link (from arXiv to IEEE).
• 2024.4.15: Upload the codes firstly.