Towards Codable Text Watermarking for Large Language Models