From 8d6a360f96b35daeb1646cccff6f73e713bf24e3 Mon Sep 17 00:00:00 2001 From: Zhou Taichang Date: Wed, 22 Oct 2025 17:13:45 +0800 Subject: [PATCH 1/2] Update vllm_omni/core/sched/diffusion_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Signed-off-by: Zhou Taichang --- vllm_omni/core/sched/diffusion_scheduler.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/vllm_omni/core/sched/diffusion_scheduler.py b/vllm_omni/core/sched/diffusion_scheduler.py index 104beec02..30ba0c4e0 100644 --- a/vllm_omni/core/sched/diffusion_scheduler.py +++ b/vllm_omni/core/sched/diffusion_scheduler.py @@ -11,9 +11,9 @@ from vllm_omni.core.sched.output import OmniNewRequestData class DiffusionScheduler(OmniScheduler): def schedule(self) -> SchedulerOutput: - """扩散快速通道： - - 一次性喂入该请求的全部输入 token（若为 0，则分配 1 个占位 token）。 - - 若无法一次性满足 token 预算，则退回上游 vLLM 的默认调度。 + """Diffusion fast path: + - Feed all input tokens of the request at once (if 0, allocate 1 placeholder token). + - If the token budget cannot be satisfied at once, fall back to the default vLLM scheduling. """ # 选出零 prompt 且使用 pooling（扩散结果经 pooler_output 回传）的请求 From b207bf89f2c63da50aff0eebb318765edc6aa485 Mon Sep 17 00:00:00 2001 From: Zhou Taichang Date: Wed, 22 Oct 2025 17:44:20 +0800 Subject: [PATCH 2/2] Update vllm_omni/core/sched/diffusion_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Signed-off-by: Zhou Taichang --- vllm_omni/core/sched/diffusion_scheduler.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vllm_omni/core/sched/diffusion_scheduler.py b/vllm_omni/core/sched/diffusion_scheduler.py index 30ba0c4e0..2c9e10c99 100644 --- a/vllm_omni/core/sched/diffusion_scheduler.py +++ b/vllm_omni/core/sched/diffusion_scheduler.py @@ -1,4 +1,6 @@ -from vllm.v1.core.sched.scheduler import SchedulerOutput, EngineCoreOutputs, Request, RequestStatus, SpecDecodingStats, defaultdict, Optional +from vllm.v1.core.sched.scheduler import SchedulerOutput, EngineCoreOutputs, Request, RequestStatus, SpecDecodingStats +from collections import defaultdict +from typing import Optional from vllm.v1.core.sched.request_queue import create_request_queue from vllm.v1.engine import EngineCoreEventType from vllm.distributed.kv_events import KVEventBatch