跳到主要内容

数据集文件格式​

对话数据集 & data_info.js​

[
  {
    "instruction": "user instruction (required)",
    "input": "user input (optional)",
    "output": "model response (required)",
    "system": "system prompt (optional)",
    "history": [
      ["user instruction in the first round (optional)", "model response in the first round (optional)"],
      ["user instruction in the second round (optional)", "model response in the second round (optional)"]
    ]
  }
]
"dataset_name": {
  "file_name": "data.json",
  "columns": {
    "prompt": "instruction",
    "query": "input",
    "response": "output",
    "system": "system",
    "history": "history"
  }
}

图像数据集 & data_info.js​

{
    "conversations": [
      {
        "from": "human",
        "value": "<image>\n请告诉我关于这张照片的整体理解，假设你是自动驾驶汽车。"
      },
      {
        "from": "gpt",
        "value": "自驾车辆正在高速直行。现场没有交通信号灯。天气晴朗。汽车正在高速公路上行驶。似乎没有行人出现。自驾车辆的驾驶员应该注意的是，要与道路上的其他车辆保持安全距离，以避免任何潜在的碰撞。"
      }
    ],
    "images": [
      "/ts-lf/ts-llamafactory/app_data/CoVLA/qa_images/images/2022-10-31--15-02-03--37_first/0002.png"
    ]
  },
"CoVLA_zh": {
    "file_name": "./CoVLA/CoVLA_llamafactory_zh_split.json",
    "formatting": "sharegpt",
    "columns": {
    "messages": "conversations",
    "images": "images"
    }
},
提示

其他类型请参考https://github.com/hiyouga/LLaMA-Factory/blob/main/data/README.md