{
  "schemaVersion": "1.0",
  "updates": [],
  "companyNews": [],
  "id": "b635",
  "language": "zh",
  "name": "SGLang",
  "originalName": "SGLang",
  "library": "tools",
  "category": "模型开发",
  "categoryKey": "模型开发",
  "categoryKeys": [
    "模型开发"
  ],
  "categories": [
    "模型开发"
  ],
  "entryPoints": [],
  "summary": "面向语言与多模态模型的推理框架，利用前缀缓存和多卡并行部署模型服务。",
  "introduction": "SGLang 是由 LMSYS 托管的开源推理框架，可部署语言、视觉语言等模型。开发者安装运行环境、加载权重并启动模型服务器，再从应用发送生成、流式输出或工具调用请求。RadixAttention 与前缀缓存用于复用共享上下文，多 GPU 并行用于扩展服务；仓库也包含图像和视频生成引擎 SGLang Diffusion。具体模型、硬件与接口支持需要按当前兼容列表核对。",
  "capabilities": [],
  "company": {
    "name": "LMSYS",
    "originalName": "LMSYS"
  },
  "officialUrl": "https://docs.sglang.io/",
  "profileUpdatedOn": "2026-10-22",
  "sources": [
    {
      "kind": "product",
      "url": "https://docs.sglang.io/"
    }
  ],
  "news": [],
  "links": {
    "interactive": "products/sglang--b635.html",
    "reading": "read/zh/products/sglang--b635.html",
    "markdown": "read/zh/products/sglang--b635.md",
    "json": "read/zh/products/sglang--b635.json"
  }
}