SGLang vs vLLM:优先级调度、限流、淘汰策略对比

📅 2026/6/29 18:07:10
SGLang vs vLLM:优先级调度、限流、淘汰策略对比
SGLang vs vLLM:优先级调度、限流、淘汰策略对比一、优先级调度维度SGLangvLLM默认策略FCFS(First Come First Serve)FCFS优先级模式--enable-priority-schedulingscheduling="priority"优先级方向默认高数值=高优先级;schedule_low_priority_values_first可反转低数值=高优先级(min-heap)排序方式(priority * sign, wait_queue_entry_time)(priority, arrival_time, request_id)