Securing LLMs using watermarking technique

This thesis focuses on implementing digital watermarking algorithms for llms and provides a analysis of how these security measures affect the quality of generated text. Using the NanoGPT architecture, the work progresses from theoretical foundations to a practical implementation and a data-driven evaluation of security strength versus model performance.