diff --git a/adapters/vllm_omni/manifest.yaml b/adapters/vllm_omni/manifest.yaml index cfbfff409..4b1cf5645 100644 --- a/adapters/vllm_omni/manifest.yaml +++ b/adapters/vllm_omni/manifest.yaml @@ -25,9 +25,9 @@ upstream: kind: fork_tracking repo_path: ${VLLM_UPSTREAM_REPO} remote: origin - # v0.28.0 alignment campaign (owner-directed 2026-08-23): wheel pick and + # v0.29.0 alignment campaign (audited 2026-09-05): wheel pick and # commit assignment target the release branch, not upstream main - target_branch: releases/v0.28.0 + target_branch: releases/v0.29.0 # Repo-specific slice of the SHARED, human-curated knowledge base (the vendored # community docs at the repo-root `knowledge/` = settings.knowledge_dir; diff --git a/adapters/vllm_omni/release_baseline.yaml b/adapters/vllm_omni/release_baseline.yaml index 717d0e566..f2fb9e247 100644 --- a/adapters/vllm_omni/release_baseline.yaml +++ b/adapters/vllm_omni/release_baseline.yaml @@ -2,9 +2,9 @@ schema_version: 1 upstream: repository: vllm-project/vllm-omni - previous_audited_sha: ae57e406fd02516ef5ee8cd40a1458328e13a2a5 - audited_ref: main - audited_sha: 1e74807c356be5cba4239faae7c38763fe1a44be + previous_audited_sha: 1e74807c356be5cba4239faae7c38763fe1a44be + audited_ref: dev/vllm-align + audited_sha: 1015b1ced073195ed19f0610264dfdf8a8150e6c # These fingerprints are generated from sorted registry/deploy inventories. # They make the baseline exact without duplicating hundreds of entries here. diff --git a/doc/architecture/KNOWLEDGE.md b/doc/architecture/KNOWLEDGE.md index 1ee4ea588..e475f4975 100644 --- a/doc/architecture/KNOWLEDGE.md +++ b/doc/architecture/KNOWLEDGE.md @@ -41,9 +41,11 @@ extensions merely because they are absent from the common source. (per-file upstream created/updated dates, captured before the submodule's git history was removed — used for page frontmatter). - **Code-mirror pin:** the `knowledge/repos/vllm-omni/components/` source maps - are verified against vllm-omni `main @ -1e74807c356be5cba4239faae7c38763fe1a44be`. The canonical machine baseline - is `adapters/vllm_omni/release_baseline.yaml`. + are verified against vllm-omni `v0.29.0-alignment @ + 1015b1ced073195ed19f0610264dfdf8a8150e6c`. Here + `v0.29.0-alignment` is a snapshot label for `dev/vllm-align`, not an + upstream tag or main revision. The canonical machine baseline is + `adapters/vllm_omni/release_baseline.yaml`. - **AFD code-mirror pin:** the `knowledge/repos/afd-plugin/components/` owner maps and architecture summaries are verified against `vllm-project/afd-plugin` default branch at diff --git a/knowledge/repos/vllm-omni/components/configuration/_index.md b/knowledge/repos/vllm-omni/components/configuration/_index.md index 12ec26f92..07ec8e561 100644 --- a/knowledge/repos/vllm-omni/components/configuration/_index.md +++ b/knowledge/repos/vllm-omni/components/configuration/_index.md @@ -15,7 +15,7 @@ sources: ["claude-workflow-starter-private@296ea45", vllm_omni/config/] `tests/diffusion/test_diffusion_config_propagation.py`，以及各公开入口附近的配置测试 - 部署配置：`vllm_omni/deploy/*.yaml`，以及 `pipeline_registry.py`、 `endpoint_policy.py`、`server_settings.py`、`yaml_util.py`、`composable_parallel/` -- 源码校验：以上路径在 `main @ 1e74807c` 验证存在；机器基线见 +- 源码校验：以上路径在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在；机器基线见 `adapters/vllm_omni/release_baseline.yaml` ## 什么时候查这里 diff --git a/knowledge/repos/vllm-omni/components/diffusion/_index.md b/knowledge/repos/vllm-omni/components/diffusion/_index.md index 143d425a7..7bbd32b56 100644 --- a/knowledge/repos/vllm-omni/components/diffusion/_index.md +++ b/knowledge/repos/vllm-omni/components/diffusion/_index.md @@ -9,8 +9,8 @@ sources: ["PR #5543", vllm_omni/platforms/cuda/platform.py, vllm_omni/diffusion/ # Diffusion -- 源码入口：`vllm_omni/diffusion/` 全树，含 16 个子模块：attention、cache、distributed、executor、hooks、layers、lora、model_loader、models、offloader、postprocess、profiler、quantization、sched、utils、worker -- 源码校验：以上子模块均已在 `main @ 1e74807c` 验证存在；MiniMax-H3 的 VAE eager +- 源码入口：`vllm_omni/diffusion/` 全树，含 17 个子模块：attention、cache、diffusion_kv、distributed、executor、hooks、layers、lora、model_loader、models、offloader、postprocess、profiler、quantization、sched、utils、worker +- 源码校验：以上子模块均已在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在；MiniMax-H3 的 VAE eager ops 仍由模型 owner 维护，其他 shared/模型机制按各自规则审查 - 主要职责：多个 diffusion 模型共用的 pipeline、执行循环、scheduler 接入和运行机制 diff --git a/knowledge/repos/vllm-omni/components/distributed/_index.md b/knowledge/repos/vllm-omni/components/distributed/_index.md index 710f03ce4..796532ad2 100644 --- a/knowledge/repos/vllm-omni/components/distributed/_index.md +++ b/knowledge/repos/vllm-omni/components/distributed/_index.md @@ -14,10 +14,10 @@ sources: [vllm_omni/distributed/omni_connectors/, vllm_omni/distributed/omni_coo `vllm_omni/distributed/omni_coordinator/`（协调器与 load balancer） - 知识面另覆盖跨 stage ZMQ 路由/端口分配（`vllm_omni/engine/stage_engine_startup.py::OmniMasterServer`） ——组件划分服务知识归属，与 manifest 运行时粒度不同 -- 源码校验：以上路径与下列锚点均已在 `main @ 1e74807c` 验证存在： +- 源码校验：以上路径与下列锚点均已在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在： `OmniConnectorBase`（connectors/base.py:12）、`OmniKVTransferManager` - （kv_transfer_manager.py:341）、`LoadBalancer` 三实现（load_balancer.py:39/64/74/102）、 - `OmniMasterServer._allocate_route_locked`（stage_engine_startup.py:254） + （kv_transfer_manager.py:348）、`LoadBalancer` 三实现（load_balancer.py:39/64/74/102）、 + `OmniMasterServer._allocate_route_locked`（stage_engine_startup.py:262） - 官方设计文档：`docs/design/feature/disaggregated_inference.md` + `docs/design/feature/omni_connectors/`（逐后端 spec） - 测试入口：`tests/distributed/` diff --git a/knowledge/repos/vllm-omni/components/model-executor/_index.md b/knowledge/repos/vllm-omni/components/model-executor/_index.md index 08b40fcd4..b3aecc212 100644 --- a/knowledge/repos/vllm-omni/components/model-executor/_index.md +++ b/knowledge/repos/vllm-omni/components/model-executor/_index.md @@ -13,7 +13,7 @@ sources: [] `vllm_omni/model_extras/`（模型专有请求/prompt 转换）、`vllm_omni/worker/`（gpu_*_worker、 gpu_*_model_runner、mixins）、`vllm_omni/inputs/`（runner 输入预处理）和设备平台层 `vllm_omni/platforms//` -- 源码校验：以上路径均已在 `main @ 1e74807c` 验证存在；stage 配置已经迁移到 +- 源码校验：以上路径均已在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在；stage 配置已经迁移到 `vllm_omni/deploy/`，NPU/XPU 等平台可继续拥有自己的 worker 覆盖 - 测试入口：共享 runner 行为看 `tests/worker/`，具体模型 consumer 看 `tests/model_executor/` - 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 diff --git a/knowledge/repos/vllm-omni/components/scheduler/_index.md b/knowledge/repos/vllm-omni/components/scheduler/_index.md index b8724dd29..7f50770af 100644 --- a/knowledge/repos/vllm-omni/components/scheduler/_index.md +++ b/knowledge/repos/vllm-omni/components/scheduler/_index.md @@ -11,9 +11,11 @@ sources: [vllm_omni/core/sched/omni_ar_scheduler.py, vllm_omni/core/sched/omni_g - 源码入口：`vllm_omni/core/sched/`（`omni_ar_scheduler.py`、`omni_generation_scheduler.py`、 `omni_scheduler_mixin.py`、`omni_scheduling_coordinator.py`）和 `vllm_omni/core/prefix_cache.py` -- 源码校验：以上路径与下列类均已在 `main @ 1e74807c` 验证存在：`OmniARScheduler`（:73）、 - `OmniARAsyncScheduler`（:815）、`OmniGenerationScheduler`（:29）、 - `OmniSchedulerMixin`（:64）、`OmniTensorPrefixCache`（prefix_cache.py:33） +- 源码校验：以上路径与下列类均已在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在： + `OmniARScheduler`（omni_ar_scheduler.py:77）、`OmniARAsyncScheduler`（:1033）、 + `KVCacheTransferData`（distributed/omni_connectors/kv_transfer_manager.py:146）、 + `OmniGenerationScheduler`（omni_generation_scheduler.py:35）、 + `OmniSchedulerMixin`（omni_scheduler_mixin.py:101）、`OmniTensorPrefixCache`（prefix_cache.py:33） - 官方 draft source map：`docs/design/module/ar_runtime.md`；旧继承关系、请求流转图只在 `docs/design/module/archive/ar_module.md` 保留，须对当前代码复核。 - 测试入口：`tests/core/` diff --git a/knowledge/repos/vllm-omni/components/serving/_index.md b/knowledge/repos/vllm-omni/components/serving/_index.md index 647b96ced..b6c1bc33d 100644 --- a/knowledge/repos/vllm-omni/components/serving/_index.md +++ b/knowledge/repos/vllm-omni/components/serving/_index.md @@ -10,7 +10,7 @@ sources: [] # Serving - 主要源码入口：`vllm_omni/entrypoints/`（cli、openai、openpi 及 omni/async_omni 入口）和 `vllm_omni/engine/`（orchestrator、stage engine core、stage pool/runtime、output processor） -- 源码校验：以上路径均已在 `main @ 1e74807c` 验证存在 +- 源码校验：以上路径均已在 `v0.29.0-alignment @ 1015b1ce`（`dev/vllm-align` 快照）验证存在 - 主要职责：用户入口、请求解析、在线服务和 engine 边界 ## 什么时候查这里 diff --git a/knowledge/repos/vllm-omni/models/catalog.md b/knowledge/repos/vllm-omni/models/catalog.md index 5e567bc80..455e6ec86 100644 --- a/knowledge/repos/vllm-omni/models/catalog.md +++ b/knowledge/repos/vllm-omni/models/catalog.md @@ -10,7 +10,8 @@ sources: [vllm_omni/model_executor/models/registry.py, vllm_omni/diffusion/regis # 模型代码入口与 registry 快照 本页提供模型描述到代码目录的自动定位入口，不维护逐模型 class 映射。下方计数是 -`main @ 1e74807c`（2026-09-05）快照，数字会漂移，不能凭它断言“不支持”。 +`v0.29.0-alignment @ 1015b1ce`（2026-09-05）快照，数字会漂移，不能凭它断言“不支持”。 +`v0.29.0-alignment` 是 `dev/vllm-align` 的快照标签，不是上游 tag 或 main 修订。 ## Direct 模型代码入口 @@ -40,6 +41,9 @@ adapter。已有专属知识 owner 可从 [models index](_index.md) 按名称进 | Pipeline（model_type） | `config/pipeline_registry.py` `OMNI_PIPELINES` | 59 个 key | | Deploy YAML | `vllm_omni/deploy/*.yaml` | 89 份 | +`1e74807c → 1015b1ce` 的对齐分支审计未改变四个 registry/deploy inventory 的 +条目或指纹；inventory 不变不代表注册逻辑或已有 deploy 文件内容没有变化。 + PR #5885 adds `MiniMaxH3TextEncoder`, the opt-in `minimax_h3_disaggregated` pipeline key, and regular/Turbo deployment profiles. It preserves bare H3's fused single-stage fallback; the split topology is selected only by its explicit deploy config. ^[PR #5885] @@ -137,8 +141,8 @@ step_audio_2_asr、voxcpm2、voxtral_tts、wan2_2_ti2v ```bash python tools/audit_vllm_omni_release.py \ - --from ae57e406 \ - --to 1e74807c \ + --from 1e74807c \ + --to 1015b1ce \ --repo \ --mode report-only ```