moezdil · Nov 3, 2026 · Nov 2, 2026 · Nov 2, 2026 · Nov 3, 2026 · Nov 3, 2026
diff --git a/docs/core-concepts/architecture.md b/docs/core-concepts/architecture.md
@@ -1,6 +1,6 @@
 ---
 title: Architecture
-description: The four HAMi components and the path a Pod takes through them.
+description: Trace an NVIDIA GPU memory request from Pod admission and card selection through device allocation to limit enforcement by HAMi-Core.
 ---
 
 The overall architecture of HAMi is shown as below:

diff --git a/docs/core-concepts/gpu-virtualization.md b/docs/core-concepts/gpu-virtualization.md
@@ -1,5 +1,6 @@
 ---
 title: GPU Virtualization Principles
+description: How HAMi selects nodes and NVIDIA GPUs with binpack or spread policies and intercepts CUDA calls to limit GPU memory and compute inside a container.
 sidebar_label: GPU Virtualization
 ---
 

diff --git a/docs/faq/faq.md b/docs/faq/faq.md
@@ -1,5 +1,6 @@
 ---
 title: FAQ
+description: Look up device support, troubleshoot GPU memory limits, and check how HAMi can coexist with the NVIDIA device plugin using distinct resource names.
 ---
 
 ## Supported Device Vendors and Specific Models

diff --git a/docs/get-started/choose-your-setup.md b/docs/get-started/choose-your-setup.md
@@ -1,5 +1,6 @@
 ---
 title: Choose Your Setup
+description: Choose classic HAMi or HAMi-DRA based on cluster requirements, or compare GPU scheduling and isolation options for Volcano and KAI Scheduler.
 sidebar_label: Choose your setup
 ---
 

diff --git a/docs/get-started/deploy-with-helm.md b/docs/get-started/deploy-with-helm.md
@@ -1,5 +1,6 @@
 ---
 title: Quick Start
+description: Install HAMi with Helm, run a sample Pod on an NVIDIA GPU, and use nvidia-smi to check how much GPU memory the container can see.
 sidebar_label: Quick Start
 ---
 

diff --git a/docs/get-started/verify-hami.md b/docs/get-started/verify-hami.md
@@ -1,5 +1,6 @@
 ---
 title: Validate HAMi Setup and vGPU Behavior
+description: Test NVIDIA GPU access with a native Pod, check the HAMi components, then run a vGPU Pod to verify its memory limit and locate deployment failures.
 sidebar_label: Validate HAMi
 ---
 

diff --git a/docs/installation/online-installation.md b/docs/installation/online-installation.md
@@ -1,6 +1,7 @@
 ---
 sidebar_label: Online installation from Helm
 title: Online installation from Helm
+description: Select a HAMi chart version, configure Helm values and image sources, then install a new release and verify the scheduler and device plugin.
 ---
 
 Install a new HAMi release with the official Helm chart. For an existing release, follow [Upgrade](./upgrade.md).

diff --git a/docs/installation/prerequisites.md b/docs/installation/prerequisites.md
@@ -1,5 +1,6 @@
 ---
 title: Prerequisites
+description: Check cluster requirements, prepare NVIDIA drivers and the container runtime, and find the setup guide for your device before installing HAMi.
 sidebar_label: Prerequisites
 ---
 

diff --git a/docs/key-features/device-sharing.md b/docs/key-features/device-sharing.md
@@ -1,6 +1,6 @@
 ---
 title: Device sharing
-description: What a Pod can ask for, and how several Pods share one card.
+description: Size NVIDIA GPU requests by device count, memory, and compute share. See how slot, memory, and compute budgets limit how many Pods fit on one card.
 ---
 
 HAMi lets multiple tasks share the same GPU, MLU, or NPU device, maximizing the utilization of heterogeneous AI computing resources.

diff --git a/docs/troubleshooting/troubleshooting.md b/docs/troubleshooting/troubleshooting.md
@@ -1,5 +1,6 @@
 ---
 title: Troubleshooting
+description: Troubleshoot HAMi device plugins and GPU workloads by symptom. Use logs, Pod events, and runtime settings to identify causes and find fixes.
 ---
 
 ## GPU Memory Limit Not Enforced {#gpu-memory-limit-not-enforced}

diff --git a/docs/userguide/device-supported.md b/docs/userguide/device-supported.md
@@ -1,5 +1,6 @@
 ---
 title: Device supported by HAMi
+description: Find supported device models, compare memory and compute isolation capabilities, and follow the setup and allocation guides for your hardware.
 displayed_sidebar: docs
 ---
 

diff --git a/docusaurus.config.js b/docusaurus.config.js
@@ -2,6 +2,8 @@ const { themes } = require("prism-react-renderer");
 const lightTheme = themes.github;
 const darkTheme = themes.dracula;
 const defaultLocale = "en";
+const llmsCatalog = require("./src/data/llms.json");
+const [latestDocsVersion] = require("./versions.json");
 const siteUrl = (process.env.DOCUSAURUS_SITE_URL || "https://project-hami.io").replace(/\/$/, "");
 const githubEditBaseUrl = "https://github.com/Project-HAMi/website/edit/master/";
 
@@ -51,7 +53,7 @@ async function localizedBlogPlugin(context, opts) {
 localizedBlogPlugin.validateOptions = require("@docusaurus/plugin-content-blog").validateOptions;
 
 /** @type {import('@docusaurus/types').DocusaurusConfig} */
-module.exports = {
+const config = {
   title: "HAMi",
   tagline: "Heterogeneous AI Computing Virtualization Middleware",
   url: siteUrl,
@@ -460,3 +462,31 @@ module.exports = {
     ],
   ],
 };
+
+module.exports = function () {
+  const locale = process.env.DOCUSAURUS_CURRENT_LOCALE || defaultLocale;
+  return {
+    ...config,
+    plugins: [
+      [
+        "docusaurus-plugin-llms",
+        {
+          description: llmsCatalog.summary[locale],
+          rootContent: llmsCatalog.guidance[locale],
+          version: latestDocsVersion,
+          docsDir: llmsCatalog.sections.map((section) => ({
+            path: `versioned_docs/version-${latestDocsVersion}/${section.path}`,
+            routeBasePath: `docs/${section.path}`,
+            label: section.title[locale],
+          })),
+          includeOrder: llmsCatalog.docs.map(
+            (id) => `versioned_docs/version-${latestDocsVersion}/${id}.md`,
+          ),
+          includeUnmatchedLast: false,
+          generateLLMsFullTxt: false,
+        },
+      ],
+      ...config.plugins,
+    ],
+  };
+};
diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/core-concepts/architecture.md b/i18n/zh/docusaurus-plugin-content-docs/current/core-concepts/architecture.md
@@ -1,6 +1,6 @@
 ---
 title: 架构设计
-description: HAMi 的四个组件，以及一个 Pod 依次经过它们的路径。
+description: 跟踪一个 Pod 的 NVIDIA GPU 显存请求，了解 Webhook 准入、调度器选卡、设备插件分配 GPU 和 HAMi-Core 在容器内执行限额的过程。
 translated: true
 ---
 

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/core-concepts/gpu-virtualization.md b/i18n/zh/docusaurus-plugin-content-docs/current/core-concepts/gpu-virtualization.md
@@ -1,5 +1,6 @@
 ---
 title: GPU 虚拟化原理
+description: HAMi 如何用 binpack 或 spread 策略选择节点和 NVIDIA GPU，并通过拦截 CUDA API 调用限制容器的显存和算力。
 sidebar_label: GPU 虚拟化
 translated: true
 ---

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/faq/faq.md b/i18n/zh/docusaurus-plugin-content-docs/current/faq/faq.md
@@ -1,5 +1,6 @@
 ---
 title: 常见问题
+description: 查询设备支持范围，排查显存限制未生效的问题，了解如何使用不同的资源名称让 HAMi 与 NVIDIA 设备插件共存。
 ---
 
 ## 支持的设备厂商及具体型号

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/get-started/choose-your-setup.md b/i18n/zh/docusaurus-plugin-content-docs/current/get-started/choose-your-setup.md
@@ -1,5 +1,6 @@
 ---
 title: 选择你的方案
+description: 根据集群条件选择经典 HAMi 或 HAMi-DRA。如果已使用 Volcano 或 KAI Scheduler，可比较它们的 GPU 调度和隔离方式。
 sidebar_label: 选择你的方案
 translated: true
 ---

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/get-started/deploy-with-helm.md b/i18n/zh/docusaurus-plugin-content-docs/current/get-started/deploy-with-helm.md
@@ -1,5 +1,6 @@
 ---
 title: 快速部署
+description: 使用 Helm 安装 HAMi，在 NVIDIA GPU 上运行示例 Pod，并用 nvidia-smi 检查容器可见的显存容量。
 sidebar_label: 快速部署
 translated: true
 ---

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/get-started/verify-hami.md b/i18n/zh/docusaurus-plugin-content-docs/current/get-started/verify-hami.md
@@ -1,5 +1,6 @@
 ---
 title: 验证 HAMi 部署与 vGPU 行为
+description: 用原生 GPU Pod 验证 NVIDIA 设备访问，检查 HAMi 组件，再用 vGPU Pod 验证显存限制，逐步定位部署故障。
 sidebar_label: 验证 HAMi
 ---
 

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/installation/online-installation.md b/i18n/zh/docusaurus-plugin-content-docs/current/installation/online-installation.md
@@ -1,6 +1,7 @@
 ---
 sidebar_label: 通过 Helm 在线安装
 title: 通过 Helm 在线安装
+description: 使用 Helm 新装 HAMi，按所选 Chart 版本准备 values 文件并配置镜像源，安装后验证调度器和设备插件。
 ---
 
 使用官方 Helm Chart 安装新的 HAMi release。已有 release 按[升级指南](./upgrade.md)操作。

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/installation/prerequisites.md b/i18n/zh/docusaurus-plugin-content-docs/current/installation/prerequisites.md
@@ -1,5 +1,6 @@
 ---
 title: 前置条件
+description: 安装 HAMi 前核对集群要求，准备 NVIDIA 节点的驱动和容器运行时，或查找其他设备的准备指南。
 sidebar_label: 前置条件
 translated: true
 ---

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/key-features/device-sharing.md b/i18n/zh/docusaurus-plugin-content-docs/current/key-features/device-sharing.md
@@ -1,6 +1,6 @@
 ---
 title: 设备共享
-description: Pod 可以申请哪些资源，以及多个 Pod 如何共享一张卡。
+description: 按 GPU 数量、显存和算力份额填写 NVIDIA 资源请求，了解槽位、显存和算力三项额度如何决定一张卡能容纳多少个 Pod。
 translated: true
 ---
 

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/troubleshooting/troubleshooting.md b/i18n/zh/docusaurus-plugin-content-docs/current/troubleshooting/troubleshooting.md
@@ -1,5 +1,6 @@
 ---
 title: 排障手册
+description: 按故障现象查找 HAMi 设备插件和 GPU 工作负载的排障步骤，结合日志、Pod 事件和运行时配置定位原因并选择修复方法。
 translated: true
 ---
 

diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/userguide/device-supported.md b/i18n/zh/docusaurus-plugin-content-docs/current/userguide/device-supported.md
@@ -1,5 +1,6 @@
 ---
 title: HAMi 支持的设备
+description: 查询 HAMi 支持的设备型号，比较各型号的显存和算力隔离能力，并查找对应的设备配置与资源分配指南。
 displayed_sidebar: docs
 translated: true
 ---

diff --git a/...zh/docusaurus-plugin-content-docs/version-v2.10.0/core-concepts/architecture.md b/...zh/docusaurus-plugin-content-docs/version-v2.10.0/core-concepts/architecture.md
@@ -1,6 +1,7 @@
 ---
 title: 架构设计
 translated: true
+description: 了解 HAMi Webhook、调度器、设备插件和 HAMi-Core 的职责，以及它们如何处理 Pod 准入、设备分配和容器内资源隔离。
 ---
 
 HAMi 的整体架构如下所示：

diff --git a/...usaurus-plugin-content-docs/version-v2.10.0/core-concepts/gpu-virtualization.md b/...usaurus-plugin-content-docs/version-v2.10.0/core-concepts/gpu-virtualization.md
@@ -2,6 +2,7 @@
 title: GPU 虚拟化原理
 sidebar_label: GPU 虚拟化
 translated: true
+description: HAMi 如何用 binpack 或 spread 策略选择节点和 NVIDIA GPU，并通过拦截 CUDA API 调用限制容器的显存和算力。
 ---
 
 在 AI 推理场景中，一个常见的困境是：GPU 很贵，但大多数时候都是闲的。

diff --git a/i18n/zh/docusaurus-plugin-content-docs/version-v2.10.0/faq/faq.md b/i18n/zh/docusaurus-plugin-content-docs/version-v2.10.0/faq/faq.md
@@ -1,5 +1,6 @@
 ---
 title: 常见问题
+description: 查询设备支持范围，排查显存限制问题，了解 vGPU 分配机制及 HAMi 与其他 Kubernetes 工具的集成关系。
 ---
 
 ## 支持的设备厂商及具体型号

diff --git a/...docusaurus-plugin-content-docs/version-v2.10.0/get-started/choose-your-setup.md b/...docusaurus-plugin-content-docs/version-v2.10.0/get-started/choose-your-setup.md
@@ -2,6 +2,7 @@
 title: 选择你的方案
 sidebar_label: 选择你的方案
 translated: true
+description: 根据集群条件选择经典 HAMi 或 HAMi-DRA。如果已使用 Volcano 或 KAI Scheduler，可比较它们的 GPU 调度和隔离方式。
 ---
 
 HAMi 有不止一种形态，旁边还有另外两个项目。本页回答大家最先问的问题：我该装哪一个？

diff --git a/.../docusaurus-plugin-content-docs/version-v2.10.0/get-started/deploy-with-helm.md b/.../docusaurus-plugin-content-docs/version-v2.10.0/get-started/deploy-with-helm.md
@@ -2,6 +2,7 @@
 title: 快速部署
 sidebar_label: 快速部署
 translated: true
+description: 使用 Helm 安装 HAMi，在 NVIDIA GPU 上运行示例 Pod，并用 nvidia-smi 检查容器可见的显存容量。
 ---
 
 本指南介绍如何使用 Helm 安装 HAMi，并以 NVIDIA GPU 为例运行 Pod、检查容器可见的显存容量。

diff --git a/i18n/zh/docusaurus-plugin-content-docs/version-v2.10.0/get-started/verify-hami.md b/i18n/zh/docusaurus-plugin-content-docs/version-v2.10.0/get-started/verify-hami.md
@@ -1,6 +1,7 @@
 ---
 title: 验证 HAMi 部署与 vGPU 行为
 sidebar_label: 验证 HAMi
+description: 用原生 GPU Pod 验证 NVIDIA 设备访问，检查 HAMi 组件，再用 vGPU Pod 验证显存限制，逐步定位部署故障。
 ---
 
 部署 HAMi 之后，验证安装是否正常工作以及 vGPU 资源隔离是否符合预期至关重要。本指南将带你逐步完成验证流程，从检查原生 GPU 栈到确认容器内的 vGPU 行为。

diff --git a/...usaurus-plugin-content-docs/version-v2.10.0/installation/online-installation.md b/...usaurus-plugin-content-docs/version-v2.10.0/installation/online-installation.md
@@ -1,6 +1,7 @@
 ---
 sidebar_label: 通过 Helm 在线安装
 title: 通过 Helm 在线安装（推荐）
+description: 通过官方 Helm Chart 安装 HAMi，自定义 values 配置，并检查调度器和设备插件的运行状态。
 ---
 
 在 Kubernetes 集群中部署 HAMi 的推荐方式是使用官方 Helm Chart。

diff --git a/...zh/docusaurus-plugin-content-docs/version-v2.10.0/installation/prerequisites.md b/...zh/docusaurus-plugin-content-docs/version-v2.10.0/installation/prerequisites.md
@@ -2,6 +2,7 @@
 title: 前置条件
 sidebar_label: 前置条件
 translated: true
+description: 安装 HAMi 前核对集群要求，准备 NVIDIA 节点的驱动和容器运行时，或查找其他设备的准备指南。
 ---
 
 安装 HAMi 前，按以下要求准备 Kubernetes 集群和设备节点。

diff --git a/...h/docusaurus-plugin-content-docs/version-v2.10.0/key-features/device-sharing.md b/...h/docusaurus-plugin-content-docs/version-v2.10.0/key-features/device-sharing.md
@@ -1,6 +1,7 @@
 ---
 title: 设备共享
 translated: true
+description: 了解 HAMi 的设备共享能力，包括显存分配、设备选择、资源限额和动态 MIG 支持。
 ---
 
 HAMi 允许多个任务共享同一块 GPU、MLU 或 NPU 设备，从而最大化异构 AI 计算资源的利用率。

diff --git a/...cusaurus-plugin-content-docs/version-v2.10.0/troubleshooting/troubleshooting.md b/...cusaurus-plugin-content-docs/version-v2.10.0/troubleshooting/troubleshooting.md
@@ -1,6 +1,7 @@
 ---
 title: 排障手册
 translated: true
+description: 按故障现象查找 HAMi 设备插件和 GPU 工作负载的排障步骤，结合日志、Pod 事件和运行时配置定位原因并选择修复方法。
 ---
 
 ## GPU 显存限制未生效 {#gpu-memory-limit-not-enforced}

diff --git a/...zh/docusaurus-plugin-content-docs/version-v2.10.0/userguide/device-supported.md b/...zh/docusaurus-plugin-content-docs/version-v2.10.0/userguide/device-supported.md
@@ -2,6 +2,7 @@
 title: HAMi 支持的设备
 displayed_sidebar: docs
 translated: true
+description: 查询 HAMi 支持的设备型号，比较各型号的显存和算力隔离能力，并查找对应的设备配置与资源分配指南。
 ---
 
 下表列出了 HAMi 支持的设备。本表反映的是 HAMi 最新发布版本 v2.10.0。

diff --git a/package-lock.json b/package-lock.json