ICX360: An Open-Source Toolkit for Context Interpretability of Large Language Models

The researchers release ICX360, an open-source Python toolkit for explaining large language model outputs, with a focus on explaining the user context or prompts given to the model. The toolkit implements three recent explanation tools, covering perturbation-based black-box methods and gradient-based white-box methods; the code is open-sourced at https://github.com/IBM/ICX360。工具包附有快速上手材料和详细教程，用例包括检索增强生成、自然语言生成和越狱。