https://www.hansking.cn/blog A posts created with Next.js and Tailwind.css en-us admin@hansking.cn (HANS) admin@hansking.cn (HANS) Sun, 12 Oct 2025 10:00:00 GMT https://www.hansking.cn/blog/2026/ContinuousBatching-yu-fu-wu-bing-fa https://www.hansking.cn/blog/2026/ContinuousBatching-yu-fu-wu-bing-fa [{"value":"Continuous Batching 与服务并发","url":"#continuous-batching-与服务并发","depth":1},{"value":"一、静态 Batch 为什么浪费","url":"#一静态-batch-为什么浪费","depth":2},{"value":"二、Continuous Batching 如何动态加入请求","url":"#二continuous-batching-如何动态加入请求","depth":2},{"value":"三、Prefill 和 Decode 是两种不同工作","url":"#三prefill-和-decode-是两种不同工作","depth":2},{"value":"四、KV Cache 是并发的硬约束","url":"#四kv-cache-是并发的硬约束","depth":2},{"value":"五、调度目标不只有吞吐","url":"#五调度目标不只有吞吐","depth":2},{"value":"六、队列和背压要尽早生效","url":"#六队列和背压要尽早生效","depth":2},{"value":"七、不同请求不要互相拖累","url":"#七不同请求不要互相拖累","depth":2},{"value":"八、流式输出不能掩盖调度问题","url":"#八流式输出不能掩盖调度问题","depth":2},{"value":"九、压测要模拟真实长度分布","url":"#九压测要模拟真实长度分布","depth":2},{"value":"总结：并发优化是在安排等待","url":"#总结并发优化是在安排等待","depth":2}] Sun, 12 Oct 2025 10:00:00 GMT admin@hansking.cn (HANS) AILLMContinuous BatchingGPU并发优化