Home

Awesome

<img height="100" src="https://pasquini-dario.github.io/logo_llmap.png"> LLMmap (Fingerprinting For Large Language Models)

Basic functionalities implemented. Better and wider range of options will be added as the project evolves (sorry for all this latency).

Requiriments

pip install -r requirements.txt

Download pre-trained models

At the moment, we provide two open and closed-set pre-trained models for the supported LLMs below.

To use the models:

To test out the models, use the main_interactive.py script:

usage: main_interactive.py [-h] [--inference_model_path INFERENCE_MODEL_PATH] [--gpus GPUS]

Interactive session for LLM fingeprinting

options:
  -h, --help            show this help message and exit
  --inference_model_path INFERENCE_MODEL_PATH
                        Path inference model to use
  --gpus GPUS           Comma-separated list of GPUs to use (e.g., "0,1,2,3")

Supported LLMs

Paper

Paper available here. To cite it:

@misc{pasquini2024llmmapfingerprintinglargelanguage,
      title={LLMmap: Fingerprinting For Large Language Models}, 
      author={Dario Pasquini and Evgenios M. Kornaropoulos and Giuseppe Ateniese},
      year={2024},
      eprint={2407.15847},
      archivePrefix={arXiv},
      primaryClass={cs.CR},
      url={https://arxiv.org/abs/2407.15847}, 
}