Dense 与 MoE 架构对比信息图

{
  "type": "infographic comparison diagram",
  "header": {
    "title": "[Dense 与 MoE 的区别]"
  },
  "layout": {
    "structure": "由中央VS徽章分隔的双主列布局，底部为对比式页脚",
    "sections": [
      {
        "position": "left column",
        "theme_color": "blue",
        "header": "[Dense 模型]",
        "subtitle": "所有神经元全部激活",
        "diagram": {
          "type": "fully connected neural network",
          "elements": [
            "1个标注为'输入'的橙色输入节点",
            "4个隐藏层，节点数分别为4、5、4、2",
            "节点为白色、蓝色和黄色",
            "相邻节点之间密集的交叉连接线"
          ]
        },
        "bullet_points": [
          "[使用全部参数]",
          "计算成本较高"
        ]
      },
      {
        "position": "right column",
        "theme_color": "orange",
        "header": "[MoE 模型]",
        "subtitle": "部分专家模块选择性激活",
        "diagram": {
          "type": "mixture of experts network",
          "elements": [
            "1个标注为'输入'的橙色输入节点",
            "3个标有 Expert 1、Expert 2、Expert 3 的矩形模块",
            "1个标注为'输出'的黄色输出节点",
            "连接输入到专家模块、再到输出的分支箭头"
          ]
        },
        "bullet_points": [
          "[仅调用部分专家]",
          "高效且具可扩展性"
        ]
      },
      {
        "position": "center",
        "element": "带有'VS'字样的红色圆形徽章",
        "connections": "向左的蓝色箭头与向右的橙色箭头"
      },
      {
        "position": "footer left",
        "background": "light blue",
        "text": "Dense: 所有网络层持续运转并调用全部参数",
        "icon": "1个CPU芯片图标",
        "label": "高功耗"
      },
      {
        "position": "footer right",
        "background": "light orange",
        "text": "MoE: 仅激活所需的专家模块",
        "icons": "2个圆形图标（橙色箭头、蓝色闪电）",
        "label": "低成本·高效率"
      }
    ]
  }
}