{
  "schemaVersion": "1.0",
  "updates": [
    {
      "id": "official-release-b634-20261018",
      "importance": "changelog",
      "product": "vLLM",
      "title": "vLLM 发布 v0.32.0",
      "summary": "增加结构化决策接口、通用解析器和更多低精度 KV 缓存支持；硬件与模型条件分别适用。",
      "originalPublishedOn": "2026-10-18",
      "source": "vLLM",
      "sourceUrl": "https://github.com/vllm-project/vllm/releases/tag/v0.32.0"
    }
  ],
  "companyNews": [],
  "id": "b634",
  "language": "zh",
  "name": "vLLM",
  "originalName": "vLLM",
  "library": "tools",
  "category": "模型开发",
  "categoryKey": "模型开发",
  "categoryKeys": [
    "模型开发"
  ],
  "categories": [
    "模型开发"
  ],
  "entryPoints": [],
  "summary": "将语言模型部署成推理服务，结合 KV 缓存管理、连续批处理、量化和多卡并行处理请求。",
  "introduction": "vLLM 是部署和运行语言模型推理的开源引擎。开发者指定模型与硬件配置后，可在进程中批量生成，也可启动兼容接口供应用发送请求。PagedAttention、连续批处理和前缀缓存用于管理推理内存与调度，框架还支持量化、结构化输出和分布式推理。它本身不提供模型权重或托管算力，实际兼容性与性能取决于模型、后端和部署配置。",
  "capabilities": [],
  "company": {
    "name": "vLLM Project",
    "originalName": "vLLM Project"
  },
  "officialUrl": "https://docs.vllm.ai/en/latest/",
  "profileUpdatedOn": "2026-10-22",
  "sources": [
    {
      "kind": "product",
      "url": "https://docs.vllm.ai/en/latest/"
    }
  ],
  "news": [],
  "links": {
    "interactive": "products/vllm--b634.html",
    "reading": "read/zh/products/vllm--b634.html",
    "markdown": "read/zh/products/vllm--b634.md",
    "json": "read/zh/products/vllm--b634.json"
  }
}