Skip to content

knowledge(vllm-omni): 1 change(s) from refine #369

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tzhouam merged 1 commit into from
Oct 13, 2026
Merged

knowledge(vllm-omni): 1 change(s) from refine #369

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions knowledge/repos/vllm-omni/components/configuration/_index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,3 +57,4 @@ sources: ["claude-workflow-starter-private@296ea45", vllm_omni/config/]

- [部署 platform overlay 的最终配置合同](rules-platform-overlays.md)。
- [配置规则 — CUDA graph 显存预留与 KV 预算](rules-cudagraph-kv-reserve.md)
- [XPU eager pins 与 Intel 支持验证](rules-xpu-eager-pins.md)
19 changes: 19 additions & 0 deletions knowledge/repos/vllm-omni/components/configuration/rules-xpu-eager-pins.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
---
title: "XPU eager pins 与 Intel 支持验证"
created: 2026-10-13
updated: 2026-10-13
type: rule
tags: [vllm-omni, components, config]
sources: ["PR #8853"]
---

# XPU eager pins 与 Intel 支持验证

## VLLM-OMNI-PR8853-XPU-EAGER-PINS — XPU 图捕获 pin 保持 stage 级 scope，经完整展开链验收

- 触发：修改 `vllm_omni/deploy/qwen3_tts.yaml`、`vllm_omni/deploy/audio8_tts.yaml` 的 `platforms.xpu` overlay（含其中的 stage 级 `enforce_eager` pin 与原因注释），或增改 `docs/models/supported_models.md` 中模型的 Intel 支持标记。
- 强制：PR #8853 在两份 YAML overlay 中建立的 pin 保持 stage 精确 scope 并与原因注释一一对应——Qwen3-TTS 仅 `stage_id: 1`（注释：Level Zero cannot capture the Code2Wav graph）、Audio8 仅 `stage_id: 0`（注释：Level Zero cannot record Fast-AR sampling）；`qwen3_tts.yaml` 的 `xpu` overlay 既有 `model_runner: v1` 等字段保持不变。触碰这些 overlay 的改动不得破坏 pin↔stage_id↔注释三者的对应关系。Intel 支持标记必须以真实 Intel 硬件与 vLLM 版本的运行记录为证据；本 PR 的记录是 Arc Pro B70、vLLM v0.31.0 上 JoyAI-Image-Edit、Qwen-Image-2.1、Kandinsky 6 零源码改动通过。
- 禁止：在 overlay 层面把 pin 扩大到其他 stage、整个模型、其他平台，或改成平台级 eager 默认；删除或改写与 stage 对应的原因注释；未在 Intel 硬件验证就添加 ✅；把这两处捕获缺陷外推为 XPU 所有 graph 不可捕获，或据此要求其他 XPU stage/模型默认 eager；以原始 YAML 单层读取代替展开后配置的验证。
- 验收：按 [CONF-3a](rules.md) 的完整链 `resolve_deploy_yaml → load_deploy_config → merge_pipeline_deploy → build_stage_runtime_overrides` 展开，断言 XPU 下 Qwen3-TTS 的 stage 1 与 Audio8 的 stage 0 在最终逐 stage 配置中生效 `enforce_eager: true`。PR #8853 的证据只到两份 YAML 的 overlay 层，未展示其他 stage 展开后的生效值；因此凡改动同 overlay 中其他 stage 或其他平台字段的 PR，必须自行给出展开后受影响 stage 的生效值。每处 pin 的原因注释仍与 stage_id 对应；新增 Intel ✅ 的模型行附硬件与 vLLM 版本的验证记录。^[PR #8853]

<!-- kb:rule status=active since=v0.30.0 -->
Loading