Python调用olmOCR大模型实现提取复杂PDF文件内容

# 安装miniconda（如果尚未安装）
wget <https://repo.anaconda.com/miniconda/miniconda3-latest-linux-x86_64.sh> -o ~/miniconda.sh ; bash ~/miniconda.sh -b -p $home/miniconda ; eval "$($home/miniconda/bin/conda shell.bash hook)" ; echo 'export path="$home/miniconda/bin:$path"' >> ~/.bashrc ; source ~/.bashrc

conda create -n ai python=3.11 -y && conda activate ai

sudo apt-get update
sudo apt-get install poppler-utils ttf-mscorefonts-installer msttcorefonts fonts-crosextra-caladea fonts-crosextra-carlito gsfonts lcdf-typetools

conda create -n olmocr python=3.11
conda activate olmocr

git clone <https://github.com/allenai/olmocr.git>
cd olmocr
pip install -e .

pip install sgl-kernel==0.0.3.post1 --force-reinstall --no-deps
pip install "sglang[all]==0.4.2" --find-links <https://flashinfer.ai/whl/cu124/torch2.4/flashinfer/>

pip install gradio pandas

python -m olmocr.pipeline ./localworkspace --pdfs tests/gnarly_pdfs/horribleocr.pdf

cat localworkspace/results/output_*.jsonl  

import os
import json
import gradio as gr
import subprocess
import pandas as pd
from pathlib import path
import shutil
import time
import re

# 创建工作目录
workspace_dir = "olmocr_workspace"
os.makedirs(workspace_dir, exist_ok=true)

def modify_html_for_better_display(html_content):
    """修改html以便在gradio中更好地显示"""
    if not html_content:
        return html_content
    
    # 增加容器宽度
    html_content = html_content.replace('<div class="container">', 
                                       '<div class="container" style="max-width: 100%; width: 100%;">')
    
    # 增加文本大小
    html_content = html_content.replace('<style>', 
                                       '<style>\nbody {font-size: 16px;}\n.text-content {font-size: 16px; line-height: 1.5;}\n')
    
    # 调整图像和文本部分的大小比例
    html_content = html_content.replace('<div class="row">', 
                                       '<div class="row" style="display: flex; flex-wrap: wrap;">')
    html_content = html_content.replace('<div class="col-md-6">', 
                                       '<div class="col-md-6" style="flex: 0 0 50%; max-width: 50%; padding: 15px;">')
    
    # 增加页面之间的间距
    html_content = html_content.replace('<div class="page">', 
                                       '<div class="page" style="margin-bottom: 30px; border-bottom: 1px solid #ccc; padding-bottom: 20px;">')
    
    # 增加图像大小
    html_content = re.sub(r'<img([^>]*)style="([^"]*)"', 
                         r'<img\1style="max-width: 100%; height: auto; \2"', 
                         html_content)
    
    # 添加缩放控制
    zoom_controls = """
    <div style="position: fixed; bottom: 20px; right: 20px; background: #fff; padding: 10px; border-radius: 5px; box-shadow: 0 0 10px rgba(0,0,0,0.2); z-index: 1000;">
        <button onclick="document.body.style.zoom = parsefloat(document.body.style.zoom || 1) + 0.1;" style="margin-right: 5px;">放大</button>
        <button onclick="document.body.style.zoom = parsefloat(document.body.style.zoom || 1) - 0.1;">缩小</button>
    </div>
    """
    html_content = html_content.replace('</body>', f'{zoom_controls}</body>')
    
    return html_content

def process_pdf(pdf_file):
    """处理pdf文件并返回结果"""
    if pdf_file is none:
        return "请上传pdf文件", "", none, none
    
    # 创建一个唯一的工作目录
    timestamp = int(time.time())
    work_dir = os.path.join(workspace_dir, f"job_{timestamp}")
    os.makedirs(work_dir, exist_ok=true)
    
    # 复制pdf文件
    pdf_path = os.path.join(work_dir, "input.pdf")
    shutil.copy(pdf_file, pdf_path)
    
    # 构建命令并执行
    cmd = ["python", "-m", "olmocr.pipeline", work_dir, "--pdfs", pdf_path]
    
    try:
        # 执行命令，等待完成
        process = subprocess.run(
            cmd,
            stdout=subprocess.pipe,
            stderr=subprocess.pipe,
            text=true,
            check=true
        )
        
        # 命令输出
        log_text = process.stdout
        
        # 检查结果目录
        results_dir = os.path.join(work_dir, "results")
        if not os.path.exists(results_dir):
            return f"处理完成，但未生成结果目录\n\n日志输出:\n{log_text}", "", none, none
        
        # 查找输出文件
        output_files = list(path(results_dir).glob("output_*.jsonl"))
        if not output_files:
            return f"处理完成，但未找到输出文件\n\n日志输出:\n{log_text}", "", none, none
        
        # 读取jsonl文件
        output_file = output_files[0]
        with open(output_file, "r") as f:
            content = f.read().strip()
            if not content:
                return f"输出文件为空\n\n日志输出:\n{log_text}", "", none, none
            
            # 解析json
            result = json.loads(content)
            extracted_text = result.get("text", "未找到文本内容")
            
            # 生成html预览
            try:
                preview_cmd = ["python", "-m", "olmocr.viewer.dolmaviewer", str(output_file)]
                subprocess.run(preview_cmd, check=true)
            except exception as e:
                log_text += f"\n生成html预览失败: {str(e)}"
            
            # 查找html文件
            html_files = list(path("dolma_previews").glob("*.html"))
            html_content = ""
            if html_files:
                try:
                    with open(html_files[0], "r", encoding="utf-8") as hf:
                        html_content = hf.read()
                        # 修改html以更好地显示
                        html_content = modify_html_for_better_display(html_content)
                except exception as e:
                    log_text += f"\n读取html预览失败: {str(e)}"
            
            # 创建元数据表格
            metadata = result.get("metadata", {})
            meta_rows = []
            for key, value in metadata.items():
                meta_rows.append([key, value])
            
            df = pd.dataframe(meta_rows, columns=["属性", "值"])
            
            return log_text, extracted_text, html_content, df
        
    except subprocess.calledprocesserror as e:
        return f"命令执行失败: {e.stderr}", "", none, none
    except exception as e:
        return f"处理过程中发生错误: {str(e)}", "", none, none

# 创建gradio界面
with gr.blocks(title="olmocr pdf提取工具") as app:
    gr.markdown("# olmocr pdf文本提取工具")
    
    with gr.row():
        with gr.column(scale=1):
            pdf_input = gr.file(label="上传pdf文件", file_types=[".pdf"])
            process_btn = gr.button("处理pdf", variant="primary")
        
        with gr.column(scale=2):
            tabs = gr.tabs()
            with tabs:
                with gr.tabitem("提取文本"):
                    text_output = gr.textbox(label="提取的文本", lines=20, interactive=true)
                with gr.tabitem("html预览", id="html_preview_tab"):
                    # 使用更大的html组件
                    html_output = gr.html(label="html预览", elem_id="html_preview_container")
                with gr.tabitem("元数据"):
                    meta_output = gr.dataframe(label="文档元数据")
                with gr.tabitem("日志"):
                    log_output = gr.textbox(label="处理日志", lines=15, interactive=false)
    
    # 使用css自定义html预览标签页和内容大小
    gr.html("""
    <style>
    #html_preview_container {
        height: 800px;
        width: 100%; 
        overflow: auto;
        border: 1px solid #ddd;
        border-radius: 4px;
    }
    #html_preview_container iframe {
        width: 100%;
        height: 100%;
        border: none;
    }
    </style>
    """)
    
    # 添加操作说明
    gr.markdown("""
    ## 使用说明
    1. 上传pdf文件
    2. 点击"处理pdf"按钮
    3. 等待处理完成
    4. 查看提取的文本和html预览
    
    ### 关于html预览
    - html预览展示原始pdf页面和提取的文本对照
    - 可以清楚地看到ocr过程的精确度
    - 如果预览内容太小，可以使用右下角的放大/缩小按钮调整
    
    ## 注意
    - 处理过程可能需要几分钟，请耐心等待
    - 首次运行会下载模型（约7gb）
    """)
    
    # 绑定按钮事件 - 使用阻塞模式
    process_btn.click(
        fn=process_pdf,
        inputs=pdf_input,
        outputs=[log_output, text_output, html_output, meta_output],
        api_name="process"
    )

# 启动应用
if __name__ == "__main__":
    app.launch(share=true)