What AI says about LLMart

LLMart is Intel Labs' open-source toolkit for evaluating the robustness of large language models (LLMs) through adversarial testing. It provides a core library plus models, datasets, and examples to attack LLMs, vision-language models, diffusion language models, and guardrail systems (e.g., HarmBench). The project adds new attack modalities, including image-based and multimodal attacks, and integrates with Intel's MART library to broaden robustness testing across model families.