Static Workload Benchmark

该数据集是为了基准测试DeServe系统的性能而设计的静态工作负载，其平均提示长度为256。数据集中包含了随机选择的长度和生成的输出。此外，该数据集还包含了在20分钟基准测试会话的最后16分钟内收集的统计数据，重点关注在变化延迟条件下的吞吐量性能。规模上，数据集模拟了同时发送N个请求的场景，任务旨在基准测试大型语言模型推理性能。

This dataset is a static workload designed for benchmarking the performance of the DeServe system, with an average prompt length of 256. It contains randomly selected prompt lengths and their generated outputs. Additionally, the dataset includes statistics collected during the final 16 minutes of a 20-minute benchmark session, focusing on throughput performance under varying latency conditions. In terms of scale, the dataset simulates a scenario where N requests are sent simultaneously, aiming to benchmark the inference performance of large language models (LLMs).