REMARK-LLM: A Robust and Efficient Watermarking Framework for Generative Large Language Models