diff --git a/knowledge/SCHEMA.md b/knowledge/SCHEMA.md index 9257bcd2..92ae37ba 100644 --- a/knowledge/SCHEMA.md +++ b/knowledge/SCHEMA.md @@ -37,7 +37,7 @@ contradictions: [相对路径] # 可选:与本页冲突的页面 - 归属:`general`、`vllm-omni`、`afd-plugin` - 工作主题:`review`、`ci`、`docs`、`git`、`debug`、`bug`、`benchmark`、`environment`、 - `remote`、`agents`、`planning`、`dev`、`rebase` + `remote`、`agents`、`planning`、`dev`、`rebase`、`release` - 代码/模型轴:`components`、`models`、`diffusion`、`model-executor`、`serving`、 `scheduler`、`distributed`、`config`、`hunyuan-image3`、`ltx2`、`qwen-omni`、 `plugin-boundary`、`attention-runtime`、`ffn-runtime`、`connectors`、 diff --git a/knowledge/general/bug/guides/di-calculation.md b/knowledge/general/bug/guides/di-calculation.md index 79acaf5f..22f8c4f5 100644 --- a/knowledge/general/bug/guides/di-calculation.md +++ b/knowledge/general/bug/guides/di-calculation.md @@ -1,7 +1,7 @@ --- title: "DI 计算说明" created: 2026-08-07 -updated: 2026-08-07 +updated: 2026-08-13 type: guide tags: [general, bug] sources: [] @@ -17,9 +17,9 @@ sources: [] | Bug 优先级分类 | 划分标准 | DI 值 | SLO | |---|---|---|---| | critical | 高优先级 | 10 | 1 天 | -| high priority | CI Failure | 3 | 3 天 | -| medium priority | 中优先级 | 1 | 5 天 | -| low priority | 低优先级 | 0.1 | 10 天 | +| high priority | CI Failure | 3 | 5 天 | +| medium priority | 中优先级 | 1 | 10 天 | +| low priority | 低优先级 | 0.1 | 14 天 | | invalid | 非问题 | 0 | / | ## 其他规则 diff --git a/knowledge/repos/vllm-omni/_index.md b/knowledge/repos/vllm-omni/_index.md index 64a87279..6c00fbbe 100644 --- a/knowledge/repos/vllm-omni/_index.md +++ b/knowledge/repos/vllm-omni/_index.md @@ -1,7 +1,7 @@ --- title: "vLLM-Omni 入口" created: 2026-07-10 -updated: 2026-07-31 +updated: 2026-08-28 type: index tags: [vllm-omni] sources: [] @@ -42,3 +42,4 @@ owner 仍不明确时才看 [components 职责表](components/_index.md);模 | benchmark / profiling | [benchmark](benchmark/_index.md) | | 远端验证 | [remote](remote/_index.md) | | 关键模型与负责人 | [owners](owners/_index.md) | +| Release 门禁与发布规范 | [rules](rules/_index.md) | diff --git a/knowledge/repos/vllm-omni/owners/_index.md b/knowledge/repos/vllm-omni/owners/_index.md index c403743b..6012e4d4 100644 --- a/knowledge/repos/vllm-omni/owners/_index.md +++ b/knowledge/repos/vllm-omni/owners/_index.md @@ -1,7 +1,7 @@ --- title: "vLLM-Omni 负责人" created: 2026-08-11 -updated: 2026-08-11 +updated: 2026-08-14 type: index tags: [vllm-omni, models] sources: [] @@ -19,9 +19,11 @@ sources: [] - 模型代码入口、registry key 和别名;这些放 [`models/catalog.md`](../models/catalog.md)。 - 单个模型的实现与运行细节;这些放 `models/<模型>/`。 +- 共享模块的职责与源码边界;这些放 [`components`](../components/_index.md)。 ## 目录内容 | 遇到什么 | 查看哪里 | 说明 | |---|---|---| | 关键模型与 Owner 清单 | [model_owner](model_owner.md) | 模型 → 负责人对照表 | +| 关键模块与 Owner 清单 | [module_owner](module_owner.md) | 模块 → 负责人对照表 | diff --git a/knowledge/repos/vllm-omni/owners/module_owner.md b/knowledge/repos/vllm-omni/owners/module_owner.md new file mode 100644 index 00000000..1249861e --- /dev/null +++ b/knowledge/repos/vllm-omni/owners/module_owner.md @@ -0,0 +1,44 @@ +--- +title: "关键模块与 Owner" +created: 2026-08-14 +updated: 2026-08-14 +type: guide +tags: [vllm-omni, components] +sources: [] +--- + +# 关键模块与 Owner + +记录 vLLM-Omni 中关键共享代码模块对应的 owner(负责人),用于 review、debug 和 +benchmark 时快速路由到正确的人。模块清单以 [components 目录](../components/_index.md) +为准,本页只维护“模块 → owner”映射。相关入口:[owners 索引](_index.md)、 +[关键模型与 Owner](model_owner.md)。 + +## 关键模块与 Owner + +| 模块 | Owner | 备注 | +|---|---|---| +| Benchmark/metrics/profiler | ZacheryAU | 区至豪 | +| Config/deploy/pipeline/cli | 外部负责 | - | +| scheduler | BLANKETusers | 张志恒 | +| coordinator | NumberWan | 温梓健 | +| Engine | fake0fan/chickeyton | 郑晨光/吴艺辉 | +| input/output processor | tzhouam | 周太昶/Shi Bo ao | +| stage_input_processor | amy-why-3459 | 吴海燕 | +| platforms | FayeSpica | 廖维明 | +| Weight loading/transformers | NumberWan | 温梓健 | +| worker/modelrunner | tzhouam | 周太昶 | +| api_server | herotai214 | 戴昊曌 | +| OmniConnector | natureofnature/spencerr221 | 刘威志/Liu Bingyu | +| Quantization | david6666666 | 陈炜青 | +| Diffusion-attention | david6666666 | 陈炜青 | +| Diffusion-cache | yangjianjuan | 杨剑娟 | +| Diffusion-distributed | bjf-frz | 白竞帆 | +| Diffusion-Executor/Scheduler/Worker/ModelRunner | fhfuih | 黄泽宇 | +| Diffusion-hooks/lora/layers/model_loader | congw729 | 王聪 | +| Diffusion-models/registry | congw729 | 王聪 | +| DiffusionEngine/StageDiffusionClient/IO | bjf-frz | 白竞帆 | +| Diffusion-model_extras | TaffyOfficial | 温智仁 | +| Diffusion-config | TaffyOfficial | 温智仁 | +| Tests | yenuo26/zhumingjue138 | 王语/朱铭觉 | + diff --git a/knowledge/repos/vllm-omni/rules/_index.md b/knowledge/repos/vllm-omni/rules/_index.md new file mode 100644 index 00000000..752d7096 --- /dev/null +++ b/knowledge/repos/vllm-omni/rules/_index.md @@ -0,0 +1,26 @@ +--- +title: "vLLM-Omni 发布规则" +created: 2026-08-28 +updated: 2026-08-28 +type: index +tags: [vllm-omni, release] +sources: [] +--- + +# vLLM-Omni 发布规则 + +## 什么时候查这里 + +- 准备发版、签收 release 门禁,或核对 CUDA/NPU/DI/需求完成度是否达标。 + +## 不放什么 + +- 仓库日常开发硬门禁;这些仍在根目录 [rules.md](../rules.md)。 +- 通用测试方法或跨仓库 CI 纪律;这些放 [general/ci](../../../general/ci/_index.md)。 +- 具体模型实现细节;这些放 [models](../models/_index.md)。 + +## 目录内容 + +| 遇到什么 | 查看哪里 | 说明 | +|---|---|---| +| Release 前必须满足的门禁 | [发布规范](release.md) | UT/性能 Guide 与 CUDA/NPU/DI 等硬门槛 | diff --git a/knowledge/repos/vllm-omni/rules/release.md b/knowledge/repos/vllm-omni/rules/release.md new file mode 100644 index 00000000..19bb875c --- /dev/null +++ b/knowledge/repos/vllm-omni/rules/release.md @@ -0,0 +1,34 @@ +--- +title: "发布规范" +created: 2026-08-28 +updated: 2026-08-28 +type: guide +tags: [vllm-omni, release] +sources: [] +--- + +# 发布规范 + +vLLM-Omni 版本 **release 前**必须满足以下门禁。标注 `(Guide)` 的项按当期迭代指南解释与举证;其余为硬性门槛。 +相关入口:[rules 索引](_index.md)、[CI](../ci/_index.md)、[DI 计算说明](../../../general/bug/guides/di-calculation.md)、[benchmark](../benchmark/_index.md)。 + +## Release 前门禁 + +| # | 要求 | 类型 | 说明 | +|---|---|---|---| +| 1 | UT coverage meets this iteration requirement | Guide | 单元测试覆盖率达到当期迭代要求;具体阈值与举证方式以当期 Guide 为准 | +| 2 | Performance regression < 10% | Guide | 性能回退小于 10%;对比基线、指标与证据口径以当期 Guide 为准 | +| 3 | Latest CUDA L1 & L2 & L3 & L4 & L5 (excluding Non-critical) pass rate = 100% | Gate | 最新一轮 CUDA L1/L2/L3/L4/L5 中,排除 Non-critical 后通过率必须为 100% | +| 4 | Latest NPU L1 & L2 & L4 pass rate = 100% | Gate | 最新一轮 NPU L2/L4 通过率必须为 100% | +| 5 | Requirement completion rate > 85% | Gate | 需求完成率大于 85% | +| 6 | Remaining DI < 30 | Gate | 剩余 DI 小于 30;DI 计算见 [DI 计算说明](../../../general/bug/guides/di-calculation.md) | +| 7 | No remaining critical issues | Gate | 无未关闭的 critical 问题 | +| 8 | All remaining bugs have assignees | Gate | 所有剩余 bug 均已指派负责人 | + +## 验收记录 + +发布前在发布说明或 checklist 中逐项给出: + +- 是否通过(Pass / Fail / N/A) +- 证据链接(CI run、benchmark 报告、需求看板、DI 统计、issue 列表) +- Guide 项写明当期 Guide 版本或链接