docker部署MinerU web api

MinerU能够将包含图片、公式、表格等元素的多模态PDF、PPT、DOCX等文档转化为易于分析的Markdown格式。

1 克隆MinerU的仓库

git clone https://github.com/opendatalab/MinerU.git

2 cd到projects/web-api

cd projects/web-api

3 在可以访问huggingface的情况下可以直接运行，后续部分步骤可能不适用于这种情况

docker build -t mineru-api .

4 如果无法通过huggingface下载，可以使用modelscope下载模型

修改requirement.txt

magic-pdf[full]==1.3.9
fastapi
uvicorn
python-multipart
requests>=2.32.3
modelscope>=1.25.0

修改download_models.py，删掉所有旧代码，改成这个

#!/usr/bin/env python
import json
import shutil
import os

import requests
from modelscope import snapshot_download


def download_json(url):
    # 下载JSON文件
    response = requests.get(url)
    response.raise_for_status()  # 检查请求是否成功
    return response.json()


def download_and_modify_json(url, local_filename, modifications):
    if os.path.exists(local_filename):
        data = json.load(open(local_filename))
        config_version = data.get('config_version', '0.0.0')
        if config_version < '1.2.0':
            data = download_json(url)
    else:
        data = download_json(url)

    # 修改内容
    for key, value in modifications.items():
        data[key] = value

    # 保存修改后的内容
    with open(local_filename, 'w', encoding='utf-8') as f:
        json.dump(data, f, ensure_ascii=False, indent=4)


if __name__ == '__main__':
    mineru_patterns = [
        # "models/Layout/LayoutLMv3/*",
        "models/Layout/YOLO/*",
        "models/MFD/YOLO/*",
        "models/MFR/unimernet_hf_small_2503/*",
        "models/OCR/paddleocr_torch/*",
        # "models/TabRec/TableMaster/*",
        # "models/TabRec/StructEqTable/*",
    ]
    model_dir = snapshot_download('opendatalab/PDF-Extract-Kit-1.0', allow_patterns=mineru_patterns)
    layoutreader_model_dir = snapshot_download('ppaanngggg/layoutreader')
    model_dir = model_dir + '/models'
    print(f'model_dir is: {model_dir}')
    print(f'layoutreader_model_dir is: {layoutreader_model_dir}')

    # paddleocr_model_dir = model_dir + '/OCR/paddleocr'
    # user_paddleocr_dir = os.path.expanduser('~/.paddleocr')
    # if os.path.exists(user_paddleocr_dir):
    #     shutil.rmtree(user_paddleocr_dir)
    # shutil.copytree(paddleocr_model_dir, user_paddleocr_dir)

    json_url = 'https://gcore.jsdelivr.net/gh/opendatalab/MinerU@master/magic-pdf.template.json'
    config_file_name = 'magic-pdf.json'
    home_dir = os.path.expanduser('~')
    config_file = os.path.join(home_dir, config_file_name)

    json_mods = {
        'models-dir': model_dir,
        'layoutreader-model-dir': layoutreader_model_dir,
    }

    download_and_modify_json(json_url, config_file, json_mods)
    print(f'The configuration file has been configured successfully, the path is: {config_file}')

5 修改Dockerfile

FROM python:3.10-slim-bookworm AS base

WORKDIR /app

ENV DEBIAN_FRONTEND=noninteractive \
    LANG=C.UTF-8 \
    PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    PIP_NO_CACHE_DIR=1


FROM base AS build

RUN echo "deb http://mirrors.aliyun.com/debian/ bookworm main non-free non-free-firmware" > /etc/apt/sources.list && \
    echo "deb http://mirrors.aliyun.com/debian-security/ bookworm-security main" >> /etc/apt/sources.list

# Update the package list and install necessary packages
RUN apt-get update && \
    apt-get install -y --no-install-recommends \
        build-essential && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

# Build Python dependencies
COPY requirements.txt .
RUN python -m venv /app/venv && \
    . /app/venv/bin/activate && \
    pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
#     pip uninstall -y paddlepaddle && \
#     pip install -i https://www.paddlepaddle.org.cn/packages/stable/cu118/ \
#         paddlepaddle-gpu==3.0.0rc1

# Download models
COPY download_models.py .
RUN . /app/venv/bin/activate && \
    ./download_models.py

FROM base AS prod

# Copy Python dependencies and models from the build stage
COPY --from=build /app/venv /app/venv
COPY --from=build /root/.cache/modelscope/hub/models/opendatalab/PDF-Extract-Kit-1___0/models /root/.cache/modelscope/hub/models/opendatalab/PDF-Extract-Kit-1___0/models
COPY --from=build /root/.cache/modelscope/hub/models/ppaanngggg/layoutreader /root/.cache/modelscope/hub/models/ppaanngggg/layoutreader
RUN echo "deb http://mirrors.aliyun.com/debian/ bookworm main non-free non-free-firmware" > /etc/apt/sources.list && \
    echo "deb http://mirrors.aliyun.com/debian-security/ bookworm-security main" >> /etc/apt/sources.list

# Update the package list and install necessary packages
RUN apt-get update && \
    apt-get install -y --no-install-recommends \
        libgl1 \
        libglib2.0-0 \
        libgomp1 && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

# Create volume for paddleocr models
# RUN mkdir -p /root/.paddleocr
# VOLUME [ "/root/.paddleocr" ]

# Copy the app and its configuration file
COPY entrypoint.sh /app/entrypoint.sh
COPY --from=build /root/magic-pdf.json /root/magic-pdf.json
COPY app.py /app/app.py

# Expose the port that FastAPI will run on
EXPOSE 8000

# Command to run FastAPI using Uvicorn, pointing to app.py and binding to 0.0.0.0:8000
ENTRYPOINT [ "/app/entrypoint.sh" ]
CMD ["--host", "0.0.0.0", "--port", "8000"]

再运行docker build -t mineru-api .

5 编写docker-compose.yml

services:
  mineru-api:
    image: mineru-api
    ports:
      - "8000:8000"
    stdin_open: true   # 对应 -i 参数
    tty: true          # 对应 -t 参数
    runtime: nvidia    # 使用 NVIDIA 容器运行时
    environment:
      - NVIDIA_VISIBLE_DEVICES=all  # 允许访问所有 GPU

6 启动

docker compose up -d

如果没有nvidia的runtime导致无法启动，先备份自己的daemon.json文件(/etc/docker/)，通过yum install -y nvidia-docker2安装，这个命令会生成一个新的daemon.json文件，覆盖掉原来旧的，安装后之后比较一下新旧两个daemon文件，合并在一起。

systemctl daemon-reload
systemctl restart docker
docker compose up -d

7 修改一下容器内的magic-pdf.json文件，使用cuda模式，默认是cpu模式

docker exec -it daa09f5404f8 /bin/sh
apt-get update
apt-get install vim
vim /root/magic-pdf.json

重启一下

docker compose restart

8 访问localhost:8000/docs

9 调用api