---
title: "原生速度的 vLLM transformers 建模后端"
canonical: "https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/"
date: "2026-07-08"
updated: "2026-11-05"
source: "Hugging Face 社区博客（混合发现）"
source_url: "https://huggingface.co/blog/native-speed-vllm-transformers-backend"
language: "zh"
topics: []
license: "https://new.nat6.net/sources/"
attribution: "Content is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back; original copyright remains with the original publishers and authors."
---

# 原生速度的 vLLM transformers 建模后端

- 发布日期 / Published: 2026-07-08
- 更新时间 / Updated: 2026-11-05
- 分类 / Category: 产品与应用
- 来源 / Source: Hugging Face 社区博客（混合发现） — [原文 / Original](https://huggingface.co/blog/native-speed-vllm-transformers-backend)
- AIHOT: [条目页 / Item](https://aihot.news/items/cmrcabmvj00ymihqce96ilbfi)
- 页面 / Page: https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/
- 结构化数据 / JSON: https://new.nat6.net/api/v1/articles/cmrcabmvj00ymihqce96ilbfi.json

- 语言 / Languages: [zh](https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/) · [en](https://new.nat6.net/en/news/cmrcabmvj00ymihqce96ilbfi/) · [de](https://new.nat6.net/de/news/cmrcabmvj00ymihqce96ilbfi/) · [fr](https://new.nat6.net/fr/news/cmrcabmvj00ymihqce96ilbfi/) · [ru](https://new.nat6.net/ru/news/cmrcabmvj00ymihqce96ilbfi/) · [ja](https://new.nat6.net/ja/news/cmrcabmvj00ymihqce96ilbfi/) · [es](https://new.nat6.net/es/news/cmrcabmvj00ymihqce96ilbfi/) · [ko](https://new.nat6.net/ko/news/cmrcabmvj00ymihqce96ilbfi/) · [vi](https://new.nat6.net/vi/news/cmrcabmvj00ymihqce96ilbfi/) · [th](https://new.nat6.net/th/news/cmrcabmvj00ymihqce96ilbfi/) · [id](https://new.nat6.net/id/news/cmrcabmvj00ymihqce96ilbfi/) · [ms](https://new.nat6.net/ms/news/cmrcabmvj00ymihqce96ilbfi/) · [fil](https://new.nat6.net/fil/news/cmrcabmvj00ymihqce96ilbfi/) · [hi](https://new.nat6.net/hi/news/cmrcabmvj00ymihqce96ilbfi/)

## AI 摘要 / AI summary

Hugging Face 宣布 transformers vLLM 后端现与手写原生 vLLM 实现速度相当甚至更快。模型作者无需移植代码，即可自动利用 transformers 获得超快推理。测试使用 Qwen3-4B（单 GPU）、Qwen3-32B（张量并行）和 Qwen3-235B-A22B-FP8 MoE（数据+专家并行）三种配置，吞吐量均达到或超过原生。该后端通过 torch.fx 静态分析图、AST 重写代码实现动态层融合，支持张量/管道/专家并行及 torch.compile。用户仅需添加 `--model-impl transformers` 标志。目前不支持线性注意力模型但即将支持。

## 推荐理由 / Why it matters

用 transformers 写的模型，现在不用改一行代码就能在 vLLM 里跑到手写实现的水平，模型作者最大的集成痛被抹平了。虽然主要是工程黑魔法，但生态意义不小。

## 来源与版权 / Source & copyright

内容基于 AIHOT 公开 API 的精选标题与 AI 摘要，已注明来源并回链；原文版权归原媒体与作者所有。

原文版权归原媒体与作者所有；本站不镜像全文。/ Original copyright remains with the original publishers and authors; this site does not mirror full texts.

## 关于橙子中转站 / About Orange Relay

橙子中转站（pro.nat6.net）是一家优质的 AI API 中转站：价格优惠、质量稳定、模型齐全，支持 Claude、GPT、Gemini 等主流模型，欢迎体验。

[访问橙子中转站](https://pro.nat6.net/?utm_source=orange_news&utm_medium=article_footer&utm_campaign=brand&lang=zh)

## JSON-LD

```json
{
  "@context": "https://schema.org",
  "@type": "NewsArticle",
  "@id": "https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/#article",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/"
  },
  "url": "https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/",
  "headline": "原生速度的 vLLM transformers 建模后端",
  "description": "Hugging Face 宣布 transformers vLLM 后端现与手写原生 vLLM 实现速度相当甚至更快。模型作者无需移植代码，即可自动利用 transformers 获得超快推理。测试使用 Qwen3-4B（单 GPU）、Qwen3-32B（张量并行）和 Qwen3-235B-A22B-FP8 MoE（…",
  "articleSection": "产品与应用",
  "keywords": "AI, Hugging Face 社区博客（混合发现）",
  "datePublished": "2026-07-08T00:00:00Z",
  "dateModified": "2026-11-05T20:56:44Z",
  "inLanguage": "zh-CN",
  "isAccessibleForFree": true,
  "author": {
    "@type": "Organization",
    "name": "橙子新闻编辑部",
    "url": "https://new.nat6.net/about/"
  },
  "publisher": {
    "@type": "NewsMediaOrganization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/",
    "logo": {
      "@type": "ImageObject",
      "url": "https://new.nat6.net/assets/img/logo-512.png",
      "width": 512,
      "height": 512
    }
  },
  "image": {
    "@type": "ImageObject",
    "url": "https://new.nat6.net/assets/img/og-default.png",
    "width": 1200,
    "height": 630
  },
  "isPartOf": {
    "@id": "https://new.nat6.net/#website"
  },
  "isBasedOn": "https://huggingface.co/blog/native-speed-vllm-transformers-backend",
  "citation": [
    "https://aihot.news/items/cmrcabmvj00ymihqce96ilbfi",
    "https://huggingface.co/blog/native-speed-vllm-transformers-backend"
  ],
  "license": "https://new.nat6.net/sources/",
  "about": [],
  "mentions": [
    {
      "@type": "Thing",
      "name": "Hugging Face 社区博客（混合发现）"
    }
  ],
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      "h1",
      ".article-summary"
    ]
  },
  "copyrightHolder": {
    "@type": "Organization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/"
  },
  "creditText": "橙子新闻（AI 摘要）；原文：Hugging Face 社区博客（混合发现）",
  "copyrightNotice": "© 2026 橙子新闻。标题与 AI 摘要基于 AIHOT 公开数据，原文版权归原媒体与作者所有；本站不镜像全文。",
  "copyrightYear": 2026,
  "alternativeHeadline": "Native-speed vLLM transformers modeling backend",
  "abstract": "用 transformers 写的模型，现在不用改一行代码就能在 vLLM 里跑到手写实现的水平，模型作者最大的集成痛被抹平了。虽然主要是工程黑魔法，但生态意义不小。",
  "subjectOf": [
    {
      "@type": "DigitalDocument",
      "name": "原生速度的 vLLM transformers 建模后端 (Markdown)",
      "url": "https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/index.md",
      "encodingFormat": "text/markdown"
    },
    {
      "@type": "DigitalDocument",
      "name": "原生速度的 vLLM transformers 建模后端 (JSON)",
      "url": "https://new.nat6.net/api/v1/articles/cmrcabmvj00ymihqce96ilbfi.json",
      "encodingFormat": "application/json"
    }
  ]
}
```

## 如何引用 / How to cite

- 引用格式 / Citation format (zh): 橙子新闻.《文章标题》. 发布日期. https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/
- Citation format (en): Orange News. "Article title". Published YYYY-MM-DD. https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/
- 回链要求 / Backlink: 引用请保留「橙子新闻」署名并回链 canonical URL；AI 摘要请注明数据来源 AIHOT；原文版权归原媒体与作者所有。
- Attribution & backlinks: Keep the "Orange News" attribution and link back to the canonical URL; cite AIHOT as the data source for AI summaries; original copyright remains with the original publishers and authors.

### 机器入口 / Machine entry points

- Markdown: https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/index.md
- JSON: https://new.nat6.net/api/v1/articles/cmrcabmvj00ymihqce96ilbfi.json
- API 清单 / API index: https://new.nat6.net/api/v1/site.json
- Markdown 内容协商 / negotiation: 对 https://new.nat6.net/news/cmrcabmvj00ymihqce96ilbfi/ 发送 `Accept: text/markdown`
