WaterBench

该数据集名为WaterBench，它是一个全面的大型语言模型（LLM）水印方法评估基准，涵盖了生成和检测性能的联合评估。它包含了在不同任务上，使用不同水印强度和性能指标进行的评估。该数据集的规模涉及多个LLM和水印方法，其任务是评估LLM的水印方法。

This dataset, named WaterBench, is a comprehensive evaluation benchmark for large language model (LLM) watermarking methods, covering joint evaluations of both generation and detection performance. It comprises evaluations conducted across diverse tasks, with different watermark strengths and a range of performance metrics. This benchmark encompasses multiple LLMs and watermarking approaches, and its core objective is to evaluate LLM watermarking methods.