I m F : Embedding an Implicit Fingerprint in Your Large Language Models
Jiaxuan Wu · Wanli Peng · Hang fu · Xue Yiming · Juan Wen
摘要
Training and serving large language models (LLMs) is resource-intensive, making reliable intellectual property (IP) protection and black-box ownership verification increasingly important.Model fingerprinting enables such verification by injecting a small set of secret query–response behaviors, bu...