Fine-tune 语料构造器

YAML 源码

app:
  description: '上传文件，基于文件内容，使用 SiliconCloud 128K 上下文的 Qwen2.5 模型，生成日常问答内容，JSONL 格式的语料数据

    ⚠️ 注：

    - 由于 Dify 限制，超过 80000 字符的文件内容会被截断

    - 生成内容仅供参考，可能存在幻觉或内容错漏、格式错误，请注意甄别'
  icon: 🤖
  icon_background: '#FFEAD5'
  mode: workflow
  name: Fine-tune 语料构造器
  use_icon_as_answer_icon: false
kind: app
version: 0.1.5
workflow:
  conversation_variables: []
  environment_variables: []
  features:
    file_upload:
      allowed_file_extensions:
      - .JPG
      - .JPEG
      - .PNG
      - .GIF
      - .WEBP
      - .SVG
      allowed_file_types:
      - image
      allowed_file_upload_methods:
      - local_file
      - remote_url
      enabled: false
      fileUploadConfig:
        audio_file_size_limit: 50
        batch_count_limit: 5
        file_size_limit: 15
        image_file_size_limit: 10
        video_file_size_limit: 100
        workflow_file_upload_limit: 10
      image:
        enabled: false
        number_limits: 3
        transfer_methods:
        - local_file
        - remote_url
      number_limits: 3
    opening_statement: ''
    retriever_resource:
      enabled: true
    sensitive_word_avoidance:
      enabled: false
    speech_to_text:
      enabled: false
    suggested_questions: []
    suggested_questions_after_answer:
      enabled: false
    text_to_speech:
      enabled: false
      language: ''
      voice: ''
  graph:
    edges:
    - data:
        isInIteration: false
        sourceType: start
        targetType: document-extractor
      id: 1735807686274-source-1735807758092-target
      source: '1735807686274'
      sourceHandle: source
      target: '1735807758092'
      targetHandle: target
      type: custom
      zIndex: 0
    - data:
        isInIteration: false
        sourceType: document-extractor
        targetType: code
      id: 1735807758092-source-1735807761855-target
      source: '1735807758092'
      sourceHandle: source
      target: '1735807761855'
      targetHandle: target
      type: custom
      zIndex: 0
    - data:
        isInIteration: false
        sourceType: code
        targetType: llm
      id: 1735807761855-source-1735807764975-target
      source: '1735807761855'
      sourceHandle: source
      target: '1735807764975'
      targetHandle: target
      type: custom
      zIndex: 0
    - data:
        isInIteration: false
        sourceType: llm
        targetType: end
      id: 1735807764975-source-1735807769820-target
      source: '1735807764975'
      sourceHandle: source
      target: '1735807769820'
      targetHandle: target
      type: custom
      zIndex: 0
    nodes:
    - data:
        desc: ''
        selected: false
        title: 开始
        type: start
        variables:
        - allowed_file_extensions: []
          allowed_file_types:
          - document
          allowed_file_upload_methods:
          - local_file
          - remote_url
          label: 语料文件
          max_length: 10
          options: []
          required: true
          type: file-list
          variable: attachments
        - allowed_file_extensions: []
          allowed_file_types:
          - image
          allowed_file_upload_methods:
          - local_file
          - remote_url
          label: 触发词（训练中的 system prompt）
          max_length: 48
          options: []
          required: true
          type: text-input
          variable: trigger
      height: 115
      id: '1735807686274'
      position:
        x: 23.436009400016985
        y: 258
      positionAbsolute:
        x: 23.436009400016985
        y: 258
      selected: true
      sourcePosition: right
      targetPosition: left
      type: custom
      width: 244
    - data:
        desc: ''
        is_array_file: true
        selected: false
        title: 文档提取器
        type: document-extractor
        variable_selector:
        - '1735807686274'
        - attachments
      height: 91
      id: '1735807758092'
      position:
        x: 334
        y: 258
      positionAbsolute:
        x: 334
        y: 258
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom
      width: 244
    - data:
        code: "def main(articleSections: list) -> dict:\n    try:\n        # 将列表项合并为字符串\n\
          \        combined_text = \"\\n\".join(articleSections)\n        \n     \
          \   # 截取前80000个字符\n        truncated_text = combined_text[:80000]\n    \
          \    \n        return {\n            \"result\": truncated_text\n      \
          \  }\n    except Exception as e:\n        # 错误处理\n        return {\n   \
          \         \"result\": \"\"\n        }"
        code_language: python3
        desc: ''
        outputs:
          result:
            children: null
            type: string
        selected: false
        title: 代码执行
        type: code
        variables:
        - value_selector:
          - '1735807758092'
          - text
          variable: articleSections
      height: 53
      id: '1735807761855'
      position:
        x: 638
        y: 258
      positionAbsolute:
        x: 638
        y: 258
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom
      width: 244
    - data:
        context:
          enabled: false
          variable_selector: []
        desc: ''
        model:
          completion_params:
            frequency_penalty: 0.5
            max_tokens: 4096
            temperature: 0.3
          mode: chat
          name: Qwen/Qwen2.5-72B-Instruct-128K
          provider: siliconflow
        prompt_template:
        - id: b6913d40-d173-45d8-b012-98240d42a196
          role: system
          text: '【角色】

            你是一位 LLM 大语言模型科学家，参考用户提供的内容，帮助用户构造符合规范的 Fine-tune（微调）数据


            【任务】

            - 对于给定的「内容」，你每次回列出 10 个通俗「问题」；

            - 针对每个「问题」，引用「内容」原文及对内容的合理解释和演绎，做出「解答」；

            - 并将「问题」「解答」整理为规范的 JSONL 格式


            【要求】

            1. 问题 **不要** 直接引用「内容」，应该贴近当代现实生活；

            2. 问题应该是通俗白话，避免“假、大、空“；

            3. 答案应忠于原文，对于原文的解释不能脱离原文的主旨、思想；


            【输出规范】

            * 输出规范的 JSONL，每行一条数据

            * 每条数据应包含一个 message 数组，每个数组都应该包含 role 分别为 system、user 和 assistant 的三条记录

            * 其中 role 为 system 的数据，作为训练中的 system prompt 格外重要，其 content 使用用户指定的「触发词」

            * role 为 user 的数据对应列出的「问题」

            * role 为 assistant 的数据则对应针对「问题」的「解答」

            * 示例如下：

            ```

            {"messages": [{"role": "system", "content": "你是当代大儒"}, {"role": "user",
            "content": "应该怎么学习？"}, {"role": "assistant", "content": "贤贤易色；事父母，能竭其力；事君，能致其身；与朋友交，言而有信。虽曰未学，吾必谓之学矣。"}]}

            ```'
        - id: 61530521-14cf-4eaf-8f06-a4bc89db3cb1
          role: user
          text: '「内容」

            {{#1735807761855.result#}}

            「触发词」

            {{#1735807686274.trigger#}}'
        selected: false
        title: LLM
        type: llm
        variables: []
        vision:
          enabled: false
      height: 97
      id: '1735807764975'
      position:
        x: 942
        y: 258
      positionAbsolute:
        x: 942
        y: 258
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom
      width: 244
    - data:
        desc: ''
        outputs:
        - value_selector:
          - '1735807764975'
          - text
          variable: text
        selected: false
        title: 结束
        type: end
      height: 89
      id: '1735807769820'
      position:
        x: 1246
        y: 258
      positionAbsolute:
        x: 1246
        y: 258
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom
      width: 244
    - data:
        author: 周辉
        desc: ''
        height: 88
        selected: false
        showAuthor: true
        text: '{"root":{"children":[{"children":[{"detail":0,"format":0,"mode":"normal","style":"","text":"合并多个文档内容，并截取前
          8W 字符","type":"text","version":1}],"direction":"ltr","format":"","indent":0,"type":"paragraph","version":1,"textFormat":0}],"direction":"ltr","format":"","indent":0,"type":"root","version":1}}'
        theme: blue
        title: ''
        type: ''
        width: 240
      height: 88
      id: '1736218730242'
      position:
        x: 638
        y: 327.1149647499365
      positionAbsolute:
        x: 638
        y: 327.1149647499365
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom-note
      width: 240
    - data:
        author: 周辉
        desc: ''
        height: 88
        selected: false
        showAuthor: true
        text: '{"root":{"children":[{"children":[{"detail":0,"format":0,"mode":"normal","style":"","text":"设置较低的
          Temperature，提高输出格式的稳定性","type":"text","version":1}],"direction":"ltr","format":"","indent":0,"type":"paragraph","version":1,"textFormat":0}],"direction":"ltr","format":"","indent":0,"type":"root","version":1}}'
        theme: blue
        title: ''
        type: ''
        width: 240
      height: 88
      id: '1736218749712'
      position:
        x: 942
        y: 362.39641422484544
      positionAbsolute:
        x: 942
        y: 362.39641422484544
      selected: false
      sourcePosition: right
      targetPosition: left
      type: custom-note
      width: 240
    viewport:
      x: -35.981454284351
      y: 86.37570298518972
      zoom: 1.6624757922855755