huawei-cloud-modelarts-vllm-ascend-model-deploy SKILL.md skillB
huawei-cloud-modelarts-vllm-ascend-model-deploy from huaweicloud/huaweicloud-skills is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add huaweicloud/huaweicloud-skills/huawei-cloud-modelarts-vllm-ascend-model-deploy@v1.0.0Pins v1.0.0 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- B, 16 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- huaweicloud/huaweicloud-skills
- Path
skills/ai/modelarts/huawei-cloud-modelarts-vllm-ascend-model-deploy/SKILL.md- Versions recorded
- 2
- Latest version
- v1.0.0, committed Oct 26, 2026
- First version recorded
- Oct 14, 2026
- Size
- 125,081 bytes
- Repository stars
- 59
- License
- MIT
指导在华为云 ModelArts 推理平台通过 vLLM-Ascend 推理框架部署开源大语言模型(GLM/Qwen/DeepSeek/Llama 等)为昇腾 NPU 在线推理服务的完整流程。覆盖:确认模型后按 vllm-ascend 社区文档与用户确认部署形式(单节点/多节点/PD分离/PD共置 4 种形式,统一单部署单元承载),从 HuggingFace/ModelScope 下载模型权重并上传 OBS、从 Quay.io 拉取 vllm-ascend 镜像并推送 SWR、按确认的部署形式选用专用脚本模板(单节点/多节点/PD/PD共置 四套,脚本内不做部署形式运行时分支)生成启动/探针/角色识别脚本并上传 OBS、按 ModelArts 平台适配要点(端口 8080、OBS 挂载、bash 数组、探针命令检查、超节点亲和组)通过新版在线服务部署(二选一: 方式A 调用 CreateInferService API(POST /v2/{project_id}/services)自动部署 / 方式B 在 ModelArts 控制台按一步步点击指导手动部署;不再创建 AI…
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# ModelArts vLLM-Ascend 开源大模型部署 ## 概述 通过 vLLM-Ascend 推理框架结合 ModelArts 推理平台的在线服务部署能力,将 HuggingFace、ModelScope 等开源社区的大语言模型(GLM、Qwen、DeepSeek、Llama 等)快速部署到昇腾 NPU 环境,提供 OpenAI 兼容的 RESTful API。本 skill 覆盖全流程: 模型确认与部署形式决策(4 种形式: 单节点/多节点/PD 分离/PD 共置,统一单部署单元)→ 权重/镜像准备 → 启动与探针脚本生成 → 新版在线服务部署(API 自动 / 控制台手动)→ OpenAI 兼容调用验证,并内置 ModelArts 平台适配要点(端口 8080、OBS 挂载、超节点亲和组、PD 分离、节点角色经 get_master_ip.py 自动判定)与安全编码规范(URL 白名单 + 用户输入校验)。 > **重要声明**: 本部署方案基于 vLLM-Ascend 开源社区提供的模型与镜像,仅适用于功能验证、原型测试等场景。生产环境的高可用、安全合规、可观测性及运维管控要求,请勿直接将本方案用于业务生产。模型能力与性能取决于开源社区。 ## 核心流程总览 **步骤 1 模型权重准备** — 确认部署模型 → 查模型文档 → 与用户确认部署形式(4 种形式 + 节点数 + 副本取舍)→ 确认权重就绪 → OBS **步骤 2 部署镜像准备** — 按模型文档+部署形式定版本/tag → 确认 SWR 就绪状态(已就绪: 记录地址 / 未就绪: 自行或代下载)→ 记录地址 **步骤 3 脚本准备** — 按确认的部署形式选用专用模板,填充模型文档参数 → 生成 `start_vllm.sh` / `health_check.sh` → **产物自检关卡(可用性硬性 + 性能软性,对照华为云最佳实践 + vllm-ascend 文档)** → OBS SDK 上传 **步骤 4 新版在线部署** — 二选一: 方式A API 自动(`create_infer_service.py` 调 CreateInferService)/ 方式B 控制台手动(9 步点击指导)—— 直接以 SWR 镜像 + OBS 挂载部署,不再创建 AI 应用(该模块已在平台废弃) **步骤 5 验证服务就绪与获取调用端点** — 查询服务状态/调用端点 → 确认 RUNNING **步骤 6 调用验证** — OpenAI 兼容 API 调用 `/v1/chat/completions` …
Read the whole file at its exact version.
How to install
mdr add huaweicloud/huaweicloud-skills/huawei-cloud-modelarts-vllm-ascend-model-deploy@sha256:ff95843bc7829e4cThe line at the top of this page pins the label v1.0.0, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_ego5epudz6xnkenm)
2 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 2 versions of this file, between Oct 14, 2026 and Oct 26, 2026, about one every 12 days.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| v1.0.0 latest | 2026-10-26 | 96691ab | 125,081 B | B | view · diff |
| v1.0.0 | 2026-10-14 | b641e81 | 96,186 B | B | view |
Audit of the latest version
- fail: No link to a raw IP address (matched: http://127.0.0.1:8080)
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (125081 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No script tag
Starting a new one? Use a template for SKILL.md on mdpantry, which passes every one of these checks and says what to change in it.
Source
huaweicloud/huaweicloud-skills · 59 stars · license MIT · pushed 2026-11-07 · branch master
API
GET https://markdownregistry.com/api/v1/artifacts/art_ego5epudz6xnkenm GET https://markdownregistry.com/api/v1/resolve?ref=huaweicloud/huaweicloud-skills/huawei-cloud-modelarts-vllm-ascend-model-deploy GET https://markdownregistry.com/api/v1/blob/ff95843bc7829e4c461fe41b8727c8362fd7d3f5f7b82af55a5ab0e96cf2d2a2
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.