dsProject/dsRagAnything/Doc/配置过程.md

一、官网

RAG-Anything 官网

二、环境配置

# 创建虚拟环境
conda create -n raganything python=3.10

# 查看当前存在哪些虚拟环境
conda env list 

# 激活虚拟环境
conda activate raganything

三、依赖环境

• 1、Libreoffice

https://zh-cn.libreoffice.org/

# 我下载的版本：
https://mirrors.nju.edu.cn/tdf/libreoffice/stable/25.2.4/win/x86_64/LibreOffice_25.2.4_Win_x86-64.msi

注：因为后面要使用的MinerU能力是将PDF转为markdown,所以需要一个将Office转成PDF的功能

下载完成后，安装即可。

• 2、RAGAnything

# 安装RagAnything
pip install raganything

# 安装包
pip install pycocotools timm
pip install detectron2 --extra-index-url https://myhloli.github.io/wheels/

# 下载模型
mineru-models-download 

四、代码调试

• 解决在国内网络无法下载huggingface的问题

# 要修改的文件
D:\anaconda3\envs\raganything\Lib\site-packages\huggingface_hub\constants.py

# 修改文件
HUGGINGFACE_CO_URL_HOME = "https://hf-mirror.com/"
_HF_DEFAULT_ENDPOINT = "https://hf-mirror.com"

• 将soffice.exe添加到环境变量

  C:\Program Files\LibreOffice\program
  

  

• 注意：需要提前配置好环境变量后，再进入PyCharm进行调试，因为我发现，如果是在打开PyCharm的前提下，添加了环境就是PyCharm里面的代码是检测不到的。

• 因为原版的程序有soffice.exe版本检测框弹出，不能直接用于生产环境，我只好手动修改了下代码：

  D:\anaconda3\envs\raganything\Lib\site-packages\raganything\mineru_parser.py
  

  修改内容：

  # Check if LibreOffice is available
              #libreoffice_available = False
              working_libreoffice_cmd = 'soffice'
              # try:
              #     result = subprocess.run(
              #         ["libreoffice", "--version"],
              #         capture_output=True,
              #         check=True,
              #         timeout=10,
              #         encoding="utf-8",
              #         errors="ignore",
              #     )
              #     libreoffice_available = True
              #     working_libreoffice_cmd = "libreoffice"
              #     print(f"LibreOffice detected: {result.stdout.strip()}")
              # except (
              #     subprocess.CalledProcessError,
              #     FileNotFoundError,
              #     subprocess.TimeoutExpired,
              # ):
              #     pass
              #
              # # Try alternative commands for LibreOffice
              # if not libreoffice_available:
              #     for cmd in ["soffice", "libreoffice"]:
              #         try:
              #             result = subprocess.run(
              #                 [cmd, "--version"],
              #                 capture_output=True,
              #                 check=True,
              #                 timeout=10,
              #                 encoding="utf-8",
              #                 errors="ignore",
              #             )
              #             libreoffice_available = True
              #             working_libreoffice_cmd = cmd
              #             print(
              #                 f"LibreOffice detected with command '{cmd}': {result.stdout.strip()}"
              #             )
              #             break
              #         except (
              #             subprocess.CalledProcessError,
              #             FileNotFoundError,
              #             subprocess.TimeoutExpired,
              #         ):
              #             continue
              #
              # if not libreoffice_available:
              #     raise RuntimeError(
              #         "LibreOffice is required for Office document conversion but was not found.\n"
              #         "Please install LibreOffice:\n"
              #         "- Windows: Download from https://www.libreoffice.org/download/download/\n"
              #         "- macOS: brew install --cask libreoffice\n"
              #         "- Ubuntu/Debian: sudo apt-get install libreoffice\n"
              #         "- CentOS/RHEL: sudo yum install libreoffice\n"
              #         "Alternatively, convert the document to PDF manually.\n"
              #         "MinerU 2.0 no longer includes built-in Office document conversion."
              #     )
  

• 首次运行时，代码会执行下面的类似命令

mineru -p C:\\Users\\ADMINI~1\\AppData\\Local\\Temp\\tmpt2sl2vd1\\驿来特平台安全.pdf -o output -m auto -b pipeline --source huggingface

下载需要等待，但程序本身不显示进度，我一直以为卡住了，后来跟踪代码，才知道它是在下载模型。

我后面修改了一下源码：

D:\anaconda3\envs\raganything\Lib\site-packages\raganything\mineru_parser.py

 # 107行
 try:
            result = subprocess.run(
                cmd,
                #capture_output=True, #注释掉这句，可以把输出打印出来
                text=True,
                check=True,
                encoding="utf-8",
                errors="ignore",
            )
            print("MinerU command executed successfully")

magic-pdf.json

{
    "bucket_info":{
        "bucket-name-1":["ak", "sk", "endpoint"],
        "bucket-name-2":["ak", "sk", "endpoint"]
    },
    "temp-output-dir":"/tmp",
    "models-dir":"C:/Users/Administrator/.cache/modelscope/hub/models/OpenDataLab/PDF-Extract-Kit-1___0/models",
    "device-mode":"cpu",
    "layout-config": {
        "model": "doclayout_yolo"
    },
    "formula-config": {
        "mfd_model": "yolo_v8_mfd",
        "mfr_model": "unimernet_small",
        "enable": true 
    },
    "table-config": {
        "model": "rapid_table",  
        "enable": false, 
        "max_time": 400
    }
}

五、相关资料

# 转换PDF到Markdown的工具
https://github.com/opendatalab/MinerU
# MinerU依赖的Magic-PDF
https://github.com/papayalove/Magic-PDF/blob/master/README_zh-CN.md
# MinerU依赖的PDF-Extract-Kit
https://github.com/opendatalab/PDF-Extract-Kit/blob/main/README_zh-CN.md

# mineru 官网
https://mineru.net/
# MinerU v2.0：VLM模型捅破解析效果天花板！
https://blog.csdn.net/qq1198768105/article/details/148678967
# MinerU、Magic-PDF、Magic-Doc
https://blog.csdn.net/lovechris00/article/details/140584728

# MinerU教程第二弹丨MinerU 本地部署保姆级“喂饭”教程
https://zhuanlan.zhihu.com/p/1908942870666282723