diff --git a/vllm/config/scheduler.py b/vllm/config/scheduler.py index 1fe09a6ae2ce3..23d21c0fd45ca 100644 --- a/vllm/config/scheduler.py +++ b/vllm/config/scheduler.py @@ -21,7 +21,11 @@ logger = init_logger(__name__) RunnerType = Literal["generate", "pooling", "draft"] SchedulerPolicy = Literal["fcfs", "priority", "sjf"] - +""" SJF Scheduling Policy: +It stands for shortest-job-first — requests are scheduled by total prompt + +output length (shorter first), with aging to prevent starvation. For more +information, please check: https://github.com/vllm-project/vllm/issues/29406 +""" @config @dataclass