What AI says about Fast-MIA

Fast-MIA is a framework for efficiently evaluating Membership Inference Attacks (MIA) against Large Language Models (LLMs). It enables fast execution of representative membership-inference methods using vLLM, with caching to preserve evaluation accuracy and the ability to compare multiple methods under the same conditions. It offers flexible configuration via YAML, supports multiple data formats (CSV, JSON, JSONL, Parquet), and can load datasets from Hugging Face Datasets.