{
  "meta": {
    "title": "AI模型训练服务 · 产品化交付包",
    "version": "1.0",
    "did": "DID-BR-000002",
    "anchor": "Ω₀⊂⊙∞⊂Ω",
    "generated": "2026-10-24",
    "note": "报价为市场默认参考(*待校准)；算力按最优稳态方案调度，轻量化工程为最高约束"
  },
  "service_overview": {
    "name": "AI模型训练托管服务",
    "value_prop": "7B/13B/34B LoRA与QLoRA全链路自治托管：数据准备→训练→评测→部署，把模型能力交给客户、训练复杂度交给服务方",
    "customers": ["想微调模型但缺infra/技能的AI团队", "企业垂直域模型", "研究者/教育机构"],
    "models": ["7B", "13B", "34B"],
    "methods": ["LoRA", "QLoRA", "全参数微调(可选)"]
  },
  "full_pipeline": {
    "steps": [
      {"step": 1, "name": "需求与数据审计", "desc": "确认目标域、评测基线、数据规模与格式", "output": "需求规格书"},
      {"step": 2, "name": "数据准备", "desc": "清洗/去重/格式化/质检/增强，构建训练集与验证集", "output": "标准数据集"},
      {"step": 3, "name": "训练执行", "desc": "LoRA/QLoRA配置与执行，多模型混合池与低成本算力调度", "output": "训练运行包+权重"},
      {"step": 4, "name": "评测", "desc": "基准评测+垂直域验证，量化提升", "output": "评测报告"},
      {"step": 5, "name": "部署", "desc": "API服务化，即调即用，含鉴权", "output": "部署服务"}
    ]
  },
  "deliverables": {
    "items": ["训练运行包(可复现)", "微调模型权重", "评测报告", "部署API服务", "技术文档+锁档SHA256"]
  },
  "pricing": {
    "tiers": [
      {"name": "7B LoRA微调", "price": "¥8-15万", "cycle": "2-4周"},
      {"name": "13B/34B LoRA/QLoRA", "price": "¥15-30万", "cycle": "4-6周"},
      {"name": "全参数微调定制", "price": "¥30万+", "cycle": "6-8周"}
    ]
  },
  "differentiators": {
    "items": [
      "最优稳态方案算力调度：多模型混合池优先、免费/低成本算力优先",
      "轻量化工程为最高约束，拒绝过度部署",
      "全链路自治托管，客户零训练团队也可交付",
      "模型权重成果归客户，交付即转移",
      "能力工程化SOP持续迭代服务质量"
    ]
  },
  "commercial_readiness": "可交付·产品矩阵P3·触达页已建(ai-training-landing.html)"
}