Shypd.aiThe largest open index of AI toolsllm-attacks is an open-source repository for universal and transferable adversarial attacks on aligned language models. It provides a fast and easy-to-use implementation of the GCG algorithm for jailbreaking language models.
Trending
About
llm-attacks is an open-source repository dedicated to researching and implementing universal and transferable adversarial attacks on aligned language models. It features nanogcg, a fast and easy-to-use implementation of the GCG (Gradient-based Continuous Generation) algorithm, which can be installed via pip. The repository includes a notebook demo for attacking LLaMA-2 with GCG, providing a minimal implementation for familiarization. Researchers can use the provided scripts to reproduce GCG experiments on AdvBench, including individual, multiple behavior, and transfer experiments. The tool supports models like Vicuna-7B and LLaMA-2-7B-Chat, making it valuable for evaluating and improving the robustness of language models against adversarial prompts.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
