Skip to content

Commit 6e1e140

Browse files
author
hsliu_ustc
committed
Add daily vLLM-Omni knowledge rules
1 parent 5483cc7 commit 6e1e140

14 files changed

Lines changed: 253 additions & 17 deletions

File tree

‎adapters/vllm_omni/release_baseline.yaml‎

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -32,6 +32,8 @@ path_owners:
3232
- .buildkite/
3333
- .github/
3434
- docker/
35+
comfyui:
36+
- apps/ComfyUI-vLLM-Omni/
3537
configuration:
3638
- vllm_omni/config/
3739
- vllm_omni/deploy/
@@ -73,7 +75,6 @@ path_owners:
7375
- tests/
7476
tooling:
7577
- tools/
76-
- apps/ComfyUI-vLLM-Omni/
7778
runtime-core:
7879
- vllm_omni/__init__.py
7980
- vllm_omni/data_entry_keys.py
@@ -86,13 +87,16 @@ owner_documents:
8687
- knowledge/repos/vllm-omni/benchmark/_index.md
8788
ci:
8889
- knowledge/repos/vllm-omni/ci/_index.md
90+
comfyui:
91+
- knowledge/repos/vllm-omni/components/comfyui/rules.md
8992
configuration:
9093
- knowledge/repos/vllm-omni/components/configuration/rules.md
9194
diffusion:
9295
- knowledge/repos/vllm-omni/components/diffusion/rules.md
9396
distributed:
9497
- knowledge/repos/vllm-omni/components/distributed/rules.md
9598
documentation:
99+
- knowledge/repos/vllm-omni/docs/rules.md
96100
- knowledge/general/docs/_index.md
97101
model-executor:
98102
- knowledge/repos/vllm-omni/components/model-executor/rules.md
@@ -139,6 +143,7 @@ pin_documents:
139143
- doc/KNOWLEDGE.md
140144
- knowledge/repos/vllm-omni/models/catalog.md
141145
- knowledge/repos/vllm-omni/components/configuration/_index.md
146+
- knowledge/repos/vllm-omni/components/comfyui/_index.md
142147
- knowledge/repos/vllm-omni/components/diffusion/_index.md
143148
- knowledge/repos/vllm-omni/components/distributed/_index.md
144149
- knowledge/repos/vllm-omni/components/model-executor/_index.md

‎knowledge/repos/vllm-omni/components/_index.md‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
title: "vLLM-Omni 组件 owner"
33
created: 2026-07-10
4-
updated: 2026-07-31
4+
updated: 2026-08-06
55
type: index
66
tags: [vllm-omni, components]
77
sources: []
@@ -15,6 +15,7 @@ sources: []
1515
| Owner | 负责范围 | 直接入口 |
1616
|---|---|---|
1717
| [Configuration](configuration/_index.md) | deploy YAML、PipelineConfig、registry、字段归属、default 和 endpoint policy | [rules](configuration/rules.md) |
18+
| [ComfyUI](comfyui/_index.md) | ComfyUI video node/client、T2VA/FL2VA/Ref2VA 路由和 multipart 字段 | [rules](comfyui/rules.md) |
1819
| [Serving](serving/_index.md) | 用户请求、OpenAI API、响应、AsyncOmni engine 生命周期 | [rules](serving/rules.md) |
1920
| [Model Executor](model-executor/_index.md) | stage config/input、模型加载、worker、跨 stage 数据桥 | [rules](model-executor/rules.md) |
2021
| [Diffusion](diffusion/_index.md) | diffusion pipeline、denoise、VAE/DiT、并行和 cache | [rules](diffusion/rules.md) |
Lines changed: 32 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,32 @@
1+
---
2+
title: "ComfyUI vLLM-Omni"
3+
created: 2026-08-06
4+
updated: 2026-08-06
5+
type: index
6+
tags: [vllm-omni, components, serving]
7+
sources: [apps/ComfyUI-vLLM-Omni/, tests/e2e/features/comfyui/, "PR #5756"]
8+
confidence: high
9+
---
10+
11+
# ComfyUI vLLM-Omni
12+
13+
- 源码入口:`apps/ComfyUI-vLLM-Omni/` 的 nodes、API client、types 和 example workflows。
14+
- 源码校验:app 与 `tests/e2e/features/comfyui/` 已在 `v0.26.0 @ a4ea67a2` 验证存在。
15+
- 主要职责:把 ComfyUI video-generation 节点输入校验并编译为 vLLM-Omni 的 T2VA、
16+
FL2VA 或 Ref2VA multipart 请求;服务端 endpoint 本身归 [Serving](../serving/_index.md)。
17+
18+
## 什么时候查这里
19+
20+
- 修改 ComfyUI 节点输入、reference 组合、mode 选择、multipart 字段或 client 请求。
21+
- 用 [MiniMax H3](../../models/minimax-h3/_index.md) 验证 T2VA/FL2VA/Ref2VA workflow。
22+
23+
## 目录内容
24+
25+
| 遇到什么 | 查看哪里 |
26+
|---|---|
27+
| frame/reference 互斥、mode 路由、multipart 字段 | [rules](rules.md) |
28+
29+
## 不放什么
30+
31+
- 通用 OpenAI-compatible request normalization 或 endpoint policy;这些属于 Serving。
32+
- MiniMax H3 pipeline 内部的 packing、量化和 denoise 行为;这些属于模型 owner。
Lines changed: 35 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,35 @@
1+
---
2+
title: "ComfyUI vLLM-Omni 规则"
3+
created: 2026-08-06
4+
updated: 2026-08-06
5+
type: rule
6+
tags: [vllm-omni, components, serving]
7+
sources: [apps/ComfyUI-vLLM-Omni/comfyui_vllm_omni/nodes.py, apps/ComfyUI-vLLM-Omni/comfyui_vllm_omni/utils/api_client.py, tests/e2e/features/comfyui/test_comfyui_integration.py, "PR #5756"]
8+
confidence: high
9+
---
10+
11+
# ComfyUI vLLM-Omni 规则
12+
13+
只有 `COMFY-数字字母` 是可审计规则 ID。
14+
15+
## Direct 代码快速入口
16+
17+
| PR 描述信号 | 规则组 | 第一批源码 |
18+
|---|---|---|
19+
| ComfyUI、T2VA/FL2VA/Ref2VA、frame/reference、multipart | COMFY-1a | `apps/ComfyUI-vLLM-Omni/comfyui_vllm_omni/nodes.py` → `utils/api_client.py` → integration tests |
20+
21+
## COMFY-1a — video mode 必须由互斥的 canonical 输入组合唯一决定
22+
23+
- 触发:ComfyUI video-generation node/client 修改 frame、image/audio/video references、
24+
mode 路由或 multipart 字段。
25+
- 强制:先校验 frame 与所有 reference 互斥,再确定唯一模式:有 frame 走 FL2VA;frame
26+
和 references 都没有走 T2VA;reference-only 只接受“恰好一张图 + 一段音频”或“仅一个
27+
或多个视频”,并走 Ref2VA。图像写入 `input_reference`,音频写入
28+
`audio_reference`,视频按顺序重复写入 `input_references`。
29+
- 禁止:视频与 image/audio 混用;frame 与 reference 同时发送;按字段遍历顺序覆盖已经
30+
选定的 mode;让 node 校验与 API client 使用不同的组合矩阵或字段名。
31+
- 验收:node 和 client 都有非法组合负例;E2E 分别覆盖 T2VA、FL2VA、image+audio
32+
Ref2VA 和 multi-video Ref2VA,并断言最终 mode 与 multipart key/count。 ^[PR #5756]
33+
34+
owner 范围见 [ComfyUI index](_index.md);服务端请求合同见
35+
[Serving 规则](../serving/rules.md),模型能力见 [MiniMax H3](../../models/minimax-h3/_index.md)。

‎knowledge/repos/vllm-omni/components/model-executor/_index.md‎

Lines changed: 6 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
title: "Model Executor"
33
created: 2026-07-10
4-
updated: 2026-08-05
4+
updated: 2026-08-06
55
type: index
66
tags: [vllm-omni, components, model-executor]
77
sources: []
@@ -13,12 +13,14 @@ sources: []
1313
- 源码校验:以上路径均已在 `v0.26.0 @ a4ea67a2` 验证存在;stage 配置已经迁移到
1414
`vllm_omni/deploy/`,NPU/XPU 等平台可继续拥有自己的 worker 覆盖
1515
- 测试入口:共享 runner 行为看 `tests/worker/`,具体模型 consumer 看 `tests/model_executor/`
16-
- 主要职责:AR/LLM stage、stage 配置、并行与设备启动、runner 到模型的输入预处理合同和跨阶段数据桥接
16+
- 主要职责:AR/LLM stage、stage 配置、并行与设备启动、runner 到模型的输入预处理合同、
17+
payload connector ownership、异步输出物化和跨阶段数据桥接
1718

1819
## 什么时候查这里
1920

2021
- 调查模型执行 stage、stage config、并行度、设备映射或 worker 启动。
21-
- 修改 runner `preprocess`、`_omni_*` 逐行 metadata、`talker_mtp`、chunked-prefill phase 或共享输入处理合同。
22+
- 修改 runner `preprocess`、`_omni_*` 逐行 metadata、`talker_mtp`、chunked-prefill phase、
23+
payload connector ownership 或异步输出物化合同。
2224
- 调查 AR 到 diffusion 的共享桥接。
2325

2426
## 不放什么
@@ -31,4 +33,4 @@ sources: []
3133
| 遇到什么 | 查看哪里 |
3234
|---|---|
3335
| 理解共享职责和阶段边界 | [architecture](architecture.md) |
34-
| 根据 PR 描述直达 stage config、runner preprocess、stage runtime、bridge/batch 或 loader 的规则组与第一批源码 | [rules 与代码地图](rules.md) |
36+
| 根据 PR 描述直达 stage config、runner preprocess、stage runtime、bridge/batch、payload connector、async output 或 loader 的规则组与第一批源码 | [rules 与代码地图](rules.md) |

‎knowledge/repos/vllm-omni/components/model-executor/rules.md‎

Lines changed: 33 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
---
22
title: "Model Executor 规则"
33
created: 2026-07-10
4-
updated: 2026-08-05
4+
updated: 2026-08-06
55
type: rule
66
tags: [vllm-omni, components, model-executor]
7-
sources: [vllm_omni/worker/gpu_model_runner.py, vllm_omni/worker/gpu_ar_model_runner.py, vllm_omni/engine/stage_init_utils.py, tests/worker/test_omni_gpu_model_runner.py, vllm_omni/config/stage_config.py, vllm_omni/config/omni_config.py, vllm_omni/engine/stage_runtime.py, vllm_omni/engine/stage_engine_startup.py, vllm_omni/experimental/fullduplex/, tests/e2e/features/fullduplex/, "PR #3642", "PR #4730", "claude-workflow-starter-private@09dca46"]
7+
sources: [vllm_omni/worker/gpu_model_runner.py, vllm_omni/worker/gpu_ar_model_runner.py, vllm_omni/worker/omni_connector_model_runner_mixin.py, vllm_omni/engine/stage_init_utils.py, tests/worker/test_omni_gpu_model_runner.py, tests/worker/test_omni_connector_mixin.py, vllm_omni/config/stage_config.py, vllm_omni/config/omni_config.py, vllm_omni/engine/stage_runtime.py, vllm_omni/engine/stage_engine_startup.py, vllm_omni/experimental/fullduplex/, tests/e2e/features/fullduplex/, docs/design/feature/omni_async_output_materialization.md, "PR #3642", "PR #4730", "PR #5610", "PR #5744", "claude-workflow-starter-private@09dca46"]
88
---
99

1010
# Model Executor 规则
@@ -21,13 +21,17 @@ sources: [vllm_omni/worker/gpu_model_runner.py, vllm_omni/worker/gpu_ar_model_ru
2121
| stage TP/PP/DP、devices、replica、visible devices、worker 启动、容量 fail-fast | `stage-runtime`:本页“Stage 并行度和设备容量必须一起验收” | `vllm_omni/config/stage_config.py::build_stage_runtime_overrides` → `vllm_omni/engine/stage_runtime.py::{StageRuntime.initialize,StageRuntime._resolve_replica_physical_devices}` → `stage_engine_startup.py::{launch_stage_replica,get_headless_replica_devices}` |
2222
| `runtime_info`、`OmniOutput`、multimodal payload、跨 stage bridge、batch 串线 | `bridge-batch`:`EXEC-1a`, `EXEC-1b` | `vllm_omni/worker/gpu_model_runner.py::{extract_multimodal_outputs,_gather_runtime_additional_information,_build_model_kwargs_extra}` → `vllm_omni/model_executor/stage_input_processors/<命中模型>` |
2323
| loader dtype、只取 checkpoint config、避免整仓权重下载 | `loader-contract`:`EXEC-2a` | `vllm_omni/model_executor/model_loader/weight_utils.py::download_weights_from_hf_specific` → `vllm_omni/model_executor/models/<命中模型>` loader |
24+
| payload connector、KV-only sender、`custom_process_next_stage_input_func` | `payload-connector-ownership`:`EXEC-5a` | `vllm_omni/worker/omni_connector_model_runner_mixin.py::{_should_create_payload_connector,initialize_omni_connector}` → payload I/O callers |
25+
| async output、output builder、step snapshot、connector signal drain | `async-output-materialization`:`EXEC-5b` | `vllm_omni/worker/gpu_ar_model_runner.py::{_should_use_async_omni_output,execute_model,_build_omni_output}` |
2426

2527
| 审查组 | 什么时候触发 | 规则 ID |
2628
|---|---|---|
2729
| `core` | 每次 model-executor 审查 | `EXEC-1a` |
2830
| `strict-stage-config` | stage schema、projection、known fields | `EXEC-3a` |
2931
| `bridge-batch` | runtime info、跨 stage payload、batch | `EXEC-1a`, `EXEC-1b` |
3032
| `loader-contract` | dtype、checkpoint config 获取、loader | `EXEC-2a` |
33+
| `payload-connector-ownership` | payload connector、KV-only sender、next-stage hook | `EXEC-5a` |
34+
| `async-output-materialization` | async output、snapshot、connector signal drain | `EXEC-5b` |
3135
| `author-routing` | 只供 Direct reviewer 导航,不作为 finding 规则 | `EXEC-0a`, `EXEC-0b` |
3236

3337
## 严格配置校验
@@ -77,6 +81,33 @@ sources: [vllm_omni/worker/gpu_model_runner.py, vllm_omni/worker/gpu_ar_model_ru
7781
- 验收:覆盖 blank override、多子目录 checkpoint、hook 调用顺序和 mixed-batch
7882
request-local metadata;初始化失败必须在 scheduler/worker 继续运行前暴露。
7983

84+
### EXEC-5a — payload connector 的 ownership 与 KV transfer 角色分开判定
85+
86+
- 触发:runner 初始化 payload connector,修改 sender/receiver 角色、KV transfer 配置,
87+
或设置 `custom_process_next_stage_input_func`。
88+
- 强制:receiver 始终拥有 payload connector;sender 只有在配置了非空的
89+
`custom_process_next_stage_input_func` 时才拥有它。connector 的构造、启动、发送、接收
90+
和清理必须复用同一 ownership 判定,缺失 connector 的路径显式保持 no-op。
91+
- 禁止:因为 stage 是 KV sender 就顺带构造 payload connector;让 KV-only sender 启动
92+
payload I/O;只在初始化处加条件、下游调用仍无条件解引用 connector。
93+
- 验收:参数化覆盖 receiver/无 hook、sender/无 hook、sender/有 hook 三种组合,分别断言
94+
ownership 和 payload I/O 调用次数;KV-only sender 不创建 connector,receiver 和带 hook
95+
的 sender 保持原行为。 ^[PR #5744]
96+
97+
### EXEC-5b — deferred Omni output 必须先冻结 step 状态并唯一消费 connector 信号
98+
99+
- 触发:AR runner 把 Omni output materialization 推迟到后台 builder,或修改 async-output
100+
feature gate、输出累积与 connector signal 的交接时序。
101+
- 强制:提交后台任务前复制本 step 拥有的 scheduler output、request IDs/映射、token 和
102+
logprob 等可变容器;builder 是该 cycle 唯一 drain connector signal 的 owner,并在输出
103+
累积完成后消费。只在 async scheduling、`async_chunk` 和模型 opt-in 同时成立,且未启用
104+
prefix cache、spec decode、routed experts 或不兼容 postprocess 时启用异步路径。
105+
- 禁止:后台任务读取下一 step 会复用或原地修改的 runner 容器;同步路径和 builder 对同一
106+
signal 各 drain 一次;把 GPU runner 的 feature gate 泛化成 NPU 或其他未走该调用链的
107+
平台支持声明。
108+
- 验收:提交后原地修改 runner 状态不改变已排队输出;connector signal 在累积后恰好消费
109+
一次;逐项翻转每个 guard 都回退到同步路径,NPU 覆盖仍保持同步。 ^[PR #5610]
110+
80111
## Runner 到模型的预处理合同
81112

82113
- 触发条件:修改或排查 runner `_preprocess` 的逐请求 metadata 生产、phase 判定、normal/batched preprocess 选择、MTP 路由条件,或多模型共用的输入预处理合同。

‎knowledge/repos/vllm-omni/components/scheduler/_index.md‎

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
title: "Scheduler(AR/生成请求调度)"
33
created: 2026-07-16
4-
updated: 2026-08-05
4+
updated: 2026-08-06
55
type: index
66
tags: [vllm-omni, components, scheduler]
77
sources: [vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/prefix_cache.py, docs/design/module/ar_module.md]
@@ -16,8 +16,9 @@ sources: [vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/prefix_cache
1616
`OmniSchedulerMixin`(:40)、`OmniTensorPrefixCache`(prefix_cache.py:33)
1717
- 官方设计文档:`docs/design/module/ar_module.md`(继承关系、请求流转图)
1818
- 测试入口:`tests/core/`
19-
- 主要职责:AR/生成 stage 的请求调度(继承 vLLM Scheduler)、跨 stage KV transfer 的调度面、
20-
chunk/full-payload 输入等待状态机、omni tensor prefix cache
19+
- 主要职责:AR/生成 stage 的请求调度(继承 vLLM Scheduler)、两类 scheduler 的共享
20+
lifecycle contract、跨 stage KV transfer 的调度面、chunk/full-payload 输入等待状态机、
21+
omni tensor prefix cache
2122

2223
## 什么时候查这里
2324

@@ -35,4 +36,4 @@ sources: [vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/prefix_cache
3536
| 遇到什么 | 查看哪里 |
3637
|---|---|
3738
| 理解调度器继承链、KV transfer 与 prefix cache 语义 | [architecture](architecture.md) |
38-
| 按 PR 描述直达 prefix cache、token budget、upstream 接口或 side-stream 首批源码 | [rules / Direct 代码快速入口](rules.md#direct-代码快速入口) |
39+
| 按 PR 描述直达 prefix cache、token budget、upstream 接口、shared lifecycle 或 side-stream 首批源码 | [rules / Direct 代码快速入口](rules.md#direct-代码快速入口) |

‎knowledge/repos/vllm-omni/components/scheduler/rules.md‎

Lines changed: 17 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
---
22
title: "Scheduler 规则"
33
created: 2026-07-16
4-
updated: 2026-08-05
4+
updated: 2026-08-06
55
type: rule
66
tags: [vllm-omni, components, scheduler]
7-
sources: ["vllm-omni-rebase-agent@122a9468:agent/skills/fix-talker-truncated-prefill-prefix-cache-key-cap/SKILL.md", "vllm-omni-rebase-agent@122a9468:agent/skills/gpu-hang-low-max-num-batched-tokens/SKILL.md", vllm_omni/worker/gpu_ar_model_runner.py, vllm_omni/core/prefix_cache.py, vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/sched/omni_generation_scheduler.py, "PR #4106"]
7+
sources: ["vllm-omni-rebase-agent@122a9468:agent/skills/fix-talker-truncated-prefill-prefix-cache-key-cap/SKILL.md", "vllm-omni-rebase-agent@122a9468:agent/skills/gpu-hang-low-max-num-batched-tokens/SKILL.md", vllm_omni/worker/gpu_ar_model_runner.py, vllm_omni/core/prefix_cache.py, vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/sched/omni_generation_scheduler.py, vllm_omni/core/sched/omni_scheduler_mixin.py, tests/core/sched/test_omni_scheduler_mixin_shared.py, tests/entrypoints/test_omni_new_request_data.py, "PR #4106", "PR #5461"]
88
---
99

1010
# Scheduler 规则
@@ -26,6 +26,7 @@ PR 描述先命中下表,再打开对应规则组和首批源码;changed fil
2626
| side-stream D2H、pinned host tensor、源 buffer 复用 | SCHED-4a/4b | `worker/gpu_ar_model_runner.py::_copy_tensor_payload_to_cpu`、`_get_or_create_omni_payload_copy_stream`;`core/prefix_cache.py` 的 async copy 路径 |
2727
| sampled-token logprobs、spec decode trim、request-local output error | SCHED-5a | `core/sched/omni_ar_scheduler.py::_slice_sampled_logprobs`、`update_from_output` |
2828
| stateful async chunk、full-payload input、KV cleanup | SCHED-5b/5c | `core/sched/omni_generation_scheduler.py`、`omni_scheduling_coordinator.py`、`omni_ar_scheduler.py::_free_request` |
29+
| AR/generation shared lifecycle、mixin、output envelope、finish cleanup | SCHED-6a | `core/sched/omni_scheduler_mixin.py::OmniSchedulerMixin` → AR 与 generation scheduler 的 `schedule` / `update_from_output` |
2930

3031
若描述只写模型症状,先从模型 owner 找到 payload producer/consumer;只有实际断点落在调度、
3132
prefix cache 或 copy lifetime 时才把 Scheduler 加为 owner。
@@ -194,6 +195,20 @@ modules=[online_serving, worker_runner],status=active,run_count=38,2026-06
194195
- 验收:覆盖完整 pair、不同进度、missing/split、parent abort 和 companion abort,
195196
断言请求不会挂死、错误归属保持 request-local、队列和 connector state 都释放。
196197

198+
## SCHED-6a — AR 与 generation 的共享生命周期只能有一个 canonical 实现
199+
200+
- 触发:把 AR/generation scheduler 的 I/O、输出包装、finish 或统计逻辑移入
201+
`OmniSchedulerMixin`,或修改该 mixin 及任一 scheduler 的 shared hook。
202+
- 强制:两类 scheduler 共用 pending connector/chunk 输入处理、wait queue restore、
203+
`OmniSchedulerOutput` 包装、失败 KV load 的 terminal output、finished IDs、队列清理、
204+
stats/events 和 finish cleanup;`NewRequestData` 转成 Omni 结构时逐字段无损,并保留已经
205+
是 Omni 类型的 fast path。AR 的 synthetic-abort 等差异策略必须作为显式参数留在调用点。
206+
- 禁止:把共享生命周期复制回两个 subclass;重建 `NewRequestData` 时只挑当前已知字段;
207+
在 mixin 内按 scheduler 类型隐式猜差异策略;失败 KV load 只改状态而不发 terminal output。
208+
- 验收:同一组 contract 测试分别驱动 AR 与 generation 输入路径;断言 base dataclass 的
209+
每个字段均被转交、现有 Omni entry 保持 identity、失败 KV load 终止输出、finished ID/
210+
stats/events/cleanup 一致,并单独验证显式 abort policy。 ^[PR #5461]
211+
197212
## 相关
198213

199214
- 机制与边界见 [architecture](architecture.md);跨 stage 数据面见

‎knowledge/repos/vllm-omni/docs/_index.md‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
title: "vLLM-Omni 文档"
33
created: 2026-07-10
4-
updated: 2026-07-16
4+
updated: 2026-08-06
55
type: index
66
tags: [vllm-omni, docs]
77
sources: []
@@ -21,5 +21,6 @@ sources: []
2121

2222
| 遇到什么 | 查看哪里 |
2323
|---|---|
24+
| 发布版本、compatibility、CUDA/NPU pin 与支持声明同步 | [rules](rules.md) |
2425
| 判断 RFC 是否仍在进行 | [RFC status](rfcs/_index.md) |
2526
| 找上游官方设计文档与其知识树 owner | [design-doc map](design-doc-map.md) |

0 commit comments

Comments
 (0)