pandora-llm 0.1.0

Red-teaming large language models for train data leakage

Pandora’s White-Box

Precise Training Data Detection and Extraction from Large Language Models

By Jeffrey G. Wang, Jason Wang, Marvin Li, and Seth Neel

Overview

pandora_llm is a red-teaming library against Large Language Models (LLMs) that assesses their vulnerability to train data leakage. It provides a unified PyTorch API for evaluating membership inference attacks (MIAs) and training data extraction.

You can read our paper and website for a technical introduction to the subject. Please refer to the documentation for the API reference as well as tutorials on how to use this codebase.

pandora_llm abides by the following core principles:

• Open Access — Ensuring that these tools are open-source for all.
• Reproducible — Committing to providing all necessary code details to ensure replicability.
• Self-Contained — Designing attacks that are self-contained, making it transparent to understand the workings of the method without having to peer through the entire codebase or unnecessary levels of abstraction, and making it easy to contribute new code.
• Model-Agnostic — Supporting any HuggingFace model and dataset, making it easy to apply to any situation.
• Usability — Prioritizing easy-to-use starter scripts and comprehensive documentation so anyone can effectively use pandora_llm regardless of prior background.

We hope that our package serves to guide LLM providers to safety-check their models before release, and to empower the public to hold them accountable to their use of data.

Installation

From source:

git clone https://github.com/safr-ai-lab/pandora-llm.git
pip install -e .

From pip:

pip install pandora-llm

Quickstart

We maintain a collection of starter scripts in our codebase under experiments/. If you are creating a new attack, we recommend making a copy of a starter script for a solid template.

python experiments/mia/run_loss.py --model_name EleutherAI/pythia-70m-deduped --model_revision step98000 --num_samples 2000 --pack --seed 229

You can reproduce the experiments described in our paper through the shell scripts provided in the scripts/ folder.

bash scripts/pretrain_mia_baselines.sh

Contributing

We welcome contributions! Please submit pull requests in our GitHub.

Citation

If you use our code or otherwise find this library useful, please cite our paper:

@article{wang2024pandora,
  title={Pandora's White-Box: Increased Training Data Leakage in Open LLMs},
  author={Wang, Jeffrey G and Wang, Jason and Li, Marvin and Neel, Seth},
  journal={arXiv preprint arXiv:2402.17012},
  year={2024}
}