Building with Pandora

pandora_llm provides a unified open framework for sharing and testing privacy attacks on LLMs.

Our philosophy is to make extending and contributing to pandora_llm easy by minimizing the effort needed to test a new attack.

Our library decomposes membership inference into the following steps:

1. Data → 2. Features → 3. Classifier → 4. Metrics

In this section, you will learn how each of these four modules work, and how they combine in the overall routine.

• Data
• Features
• Classifiers
• Metrics
• Models
• Routines