Dense 與 MoE 架構對比資訊圖

{
  "type": "infographic comparison diagram",
  "header": {
    "title": "[Dense 與 MoE 的區別]"
  },
  "layout": {
    "structure": "由中央VS徽章分隔的雙主列布局，底部為對比式頁尾",
    "sections": [
      {
        "position": "left column",
        "theme_color": "blue",
        "header": "[Dense 模型]",
        "subtitle": "所有神經元全部啟用",
        "diagram": {
          "type": "fully connected neural network",
          "elements": [
            "1個標註為'輸入'的橙色輸入節點",
            "4個隱藏層，節點數分別為4、5、4、2",
            "節點為白色、藍色和黃色",
            "相鄰節點之間密集的交叉連線線"
          ]
        },
        "bullet_points": [
          "[使用全部引數]",
          "計算成本較高"
        ]
      },
      {
        "position": "right column",
        "theme_color": "orange",
        "header": "[MoE 模型]",
        "subtitle": "部分專家模組選擇性啟用",
        "diagram": {
          "type": "mixture of experts network",
          "elements": [
            "1個標註為'輸入'的橙色輸入節點",
            "3個標有 Expert 1、Expert 2、Expert 3 的矩形模組",
            "1個標註為'輸出'的黃色輸出節點",
            "連線輸入到專家模組、再到輸出的分支箭頭"
          ]
        },
        "bullet_points": [
          "[僅呼叫部分專家]",
          "高效且具可擴充套件性"
        ]
      },
      {
        "position": "center",
        "element": "帶有'VS'字樣的紅色圓形徽章",
        "connections": "向左的藍色箭頭與向右的橙色箭頭"
      },
      {
        "position": "footer left",
        "background": "light blue",
        "text": "Dense: 所有網路層持續運轉並呼叫全部引數",
        "icon": "1個CPU晶片圖示",
        "label": "高功耗"
      },
      {
        "position": "footer right",
        "background": "light orange",
        "text": "MoE: 僅啟用所需的專家模組",
        "icons": "2個圓形圖示（橙色箭頭、藍色閃電）",
        "label": "低成本·高效率"
      }
    ]
  }
}