llm-deploy-training SKILL.md skillA
llm-deploy-training from black-yt/skills is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add black-yt/skills/llm-deploy-training@git:20260804.1ce2bfePins git:20260804.1ce2bfe in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- black-yt/skills
- Path
llm-deploy-training/SKILL.md- Versions recorded
- 5
- Latest version
- git:20260804.1ce2bfe, committed Aug 4, 2026
- First version recorded
- Jun 11, 2026
- Size
- 10,526 bytes
- Repository stars
- 16
- License
- none declared
当需要部署或训练 LLM/VLM 时使用；覆盖 vLLM OpenAI-compatible 服务、多模态输入限制、Qwen3.5 工具调用、thinking/reasoning 控制、CUDA Graph 策略，以及 ms-swift SFT/DPO/GRPO full training、Megatron 长序列训练、messages loss/loss_scale、数据校验、显存排错和训练检查。.
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
# LLM Deploy And Training ## 文件导航 | 序号 | 文件内容概览 | 关键词 | 触发时机 | 文件路径 | | --- | --- | --- | --- | --- | | 1 | 规定 LLM/VLM 部署和训练的共通安全边界，说明不要改共享环境、如何匹配 vLLM/ms-swift/transformers/torch 版本、何时读官方文档、何时只读追溯源码，并记录 ms-swift 官方文档入口。 | LLM deploy、VLM deploy、training、shared env、conda、version match、official docs、CLI help、source tracing、site-packages、vLLM、ms-swift、loss、loss_scale、transformers、torch | 触发本 skill 后默认读取；部署或训练前；准备安装/升级依赖前；参数不确定要查文档或源码前；涉及共享 conda、CUDA、torch/vLLM/ms-swift 环境时读取 | `SKILL.md` | | 2 | 记录 vLLM OpenAI-compatible 服务的完整部署和验证经验，覆盖 `vllm serve` 参数、多模态 `limit-mm-per-prompt`、Qwen3.5 工具调用、thinking/reasoning、35B 2 卡、CUDA Graph、`extra_body`、内网访问和 SSH local port forwarding。 | vLLM、OpenAI-compatible server、`vllm serve`、multimodal、`limit-mm-per-prompt`、auto tool choice、`qwen3_coder`、Qwen thinking、reasoning parser、35B、tensor parallel、CUDA Graph、`extra_body`、SSH forwarding、`/v1/models` | 写或审查 vLLM 部署脚本前；验证 `/v1/models`/短文本/工具调用/图片输入前；配置 Qwen thinking 开关或 35B 服务前；排查 CUDA Graph、OOM、多模态报错、工具调用失败或本地访问内网服务时必须读取 | [references/vllm-deployment.md](references/vllm-deployment.md) | …
Read the whole file at its exact version.
How to install
mdr add black-yt/skills/llm-deploy-training@sha256:b3a248cc0485bb5eThe line at the top of this page pins the label git:20260804.1ce2bfe, 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_l4nzgiktbwzelyy7)
3 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 5 versions of this file, between Jun 11, 2026 and Aug 4, 2026, about one every 13 days.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20260804.1ce2bfe latest | 2026-08-04 | 1ce2bfe | 10,526 B | A | view · diff |
| git:20260723.84300d7 | 2026-07-23 | 84300d7 | 9,176 B | A | view · diff |
| git:20260723.969adcb | 2026-07-23 | 969adcb | 8,595 B | A | view · diff |
| git:20260721.f8b0b02 | 2026-07-21 | f8b0b02 | 8,536 B | A | view · diff |
| git:20260611.295e62a | 2026-06-11 | 295e62a | 6,726 B | A | view |
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (10526 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 link to a raw IP address
- pass: No script tag
Source
black-yt/skills · 16 stars · license none declared · pushed 2026-11-05 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_l4nzgiktbwzelyy7 GET https://markdownregistry.com/api/v1/resolve?ref=black-yt/skills/llm-deploy-training GET https://markdownregistry.com/api/v1/blob/b3a248cc0485bb5ecd9bdebec15eeed6e082575a6179845f55c6f74ca78227b4
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.