DeepSeek-R1-Distill-Qwen模型性能Benchmark

测试条件

  • 测试开发板:S100P。

  • 性能数据获取:测试单条prompt,取TTFT(首token延迟)和TPS(平均每秒Token数)指标。

  • Python版本:Python3.10。

  • 运行环境:Linux。

实测数据

modelplatformdtypeseqlenmax contextTTFT(ms)TPSmemory(GB)
DeepSeek-R1-Distill-Qwen-1.5BS100Pq8256102410927.081.7
DeepSeek-R1-Distill-Qwen-1.5BS100Pq4256102410839.491.1
DeepSeek-R1-Distill-Qwen-1.5BS100Pq8256409622623.801.8
DeepSeek-R1-Distill-Qwen-1.5BS100Pq4256409622432.351.2
DeepSeek-R1-Distill-Qwen-7BS100Pq825610245446.767.4