flâneur

SoyGema/pulling_ace

github.com · 546 words · saved by 1 readers

PullingAce is a Python library designed to benchmark adversarial attacks on Hugging Face models. Built on top of TextAttack and Garak, PullingAce incorporates a set of recipes and attacks to assess the robustness of various natural language processing models in classification and generation taks. This tool provides a comprehensive evaluation of model vulnerabilities and helps researchers and practitioners in the field of machine learning understand the strengths and weaknesses of different models. Adversarial Attack Benchmarks: PullingAce provides a collection of adversarial attack benchmarks tailored for Hugging Face models. Evaluate model robustness against state-of-the-art attacks. Incorporating TextAttack Recipes: PullingAce integrates TextAttack's powerful attack recipes, making it easy to experiment with different attack strategies and customize evaluations. Prompt Injection: PullingAce integrates the prompt injection feature from the Garak library, allowing for more dynamic and

PullingAce: Benchmarking Robustness for LLM Models Note Repository under active construction PullingAce.mp4 PullingAce is a Python library designed to benchmark adversarial attacks on Hugging Face models. Built on top of TextAttack and Garak, PullingAce incorporates a set of recipes and attacks to assess the robustness of various natural language processing models in classification and generation taks. This tool provides a comprehensive evaluation of model vulnerabilities and helps researchers and practitioners in the field of machine learning understand the strengths and weaknesses of…

saved by

related reading