LineageMark: Multi-user White-box Watermarking for Contribution Tracing in Model Derivation Chains

LineageMark is a newly proposed multi-user white-box watermarking framework designed for tracking contributions in model derivation chains of large language models (LLMs). It utilizes a projection-based approach to embed watermarks in model parameters, ensuring robustness against changes and perturbations during incremental updates, re-watermarking, fine-tuning, quantization, and pruning. This framework is significant for practitioners as it facilitates reliable provenance tracking and intellectual property protection in collaborative LLM development environments.