Break the GPU barrier! This project enables you to run PaddleOCR-VL (PaddlePaddle's excellent OCR model) on CPU with an OpenAI-compatible API interface.

• 🖥️ CPU-Friendly - Optimized for CPU inference with automatic memory management
• 🔌 OpenAI-Compatible - Drop-in replacement for OpenAI's vision API
• 🚀 Easy Deployment - One command to install and start
• ⚡ Streaming Support - Real-time response streaming
• 💾 Auto Memory Cleanup - Prevents memory leaks for long-running services
• 🖼️ Multiple Formats - Supports JPEG, PNG, BMP, Base64, and URLs

• Python 3.12.3
• 16GB+ RAM (recommended for CPU inference)
• Ubuntu/Linux (Windows users please use WSL)

# Clone repository
git clone https://github.com/Think-Core/paddleocr-vl-cpu.git
cd paddleocr-vl-cpu

# One-click setup (downloads model automatically ~2GB)
chmod +x quick_init_setup.sh
./quick_init_setup.sh

The script will:

1. ✅ Create Python virtual environment
2. ✅ Install all dependencies
3. ✅ Download PaddleOCR-VL-0.9B model (~2GB)
4. ✅ Verify installation

# Start server
./quick_start.sh start

# Check status
./quick_start.sh status

# View logs
./quick_start.sh attach  # Press Ctrl+B then D to detach

# Stop server
./quick_start.sh stop

Server runs on http://localhost:7777

import requests
import base64

# Read and encode image
with open("document.jpg", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode()

# OCR request
response = requests.post(
    "http://localhost:7777/v1/chat/completions",
    json={
        "model": "paddleocr-vl",
        "messages": [{
            "role": "user",
            "content": [
                {"type": "text", "text": "Extract all text from this image"},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": f"data:image/jpeg;base64,{image_b64}"
                    }
                }
            ]
        }],
        "max_tokens": 2048
    }
)

print(response.json()["choices"][0]["message"]["content"])

import json

response = requests.post(
    "http://localhost:7777/v1/chat/completions",
    json={
        "model": "paddleocr-vl",
        "messages": [...],  # same as above
        "stream": True
    },
    stream=True
)

for line in response.iter_lines():
    if line:
        line = line.decode('utf-8')
        if line.startswith('data: '):
            data = line[6:]
            if data != '[DONE]':
                chunk = json.loads(data)
                content = chunk["choices"][0]["delta"].get("content", "")
                print(content, end='', flush=True)

curl -X POST http://localhost:7777/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "paddleocr-vl",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Read this document"},
        {
          "type": "image_url",
          "image_url": {"url": "data:image/jpeg;base64,{image_b64}"}
        }
      ]
    }]
  }'

• 📄 Document Digitization - Convert scanned documents to text
• 🧾 Invoice Processing - Extract invoice information automatically
• 📋 Form Recognition - Parse structured forms and tables
• 🔍 ID Card Reading - Extract text from ID cards and certificates
• 📚 Book Scanning - Digitize printed books and articles
• 📝 Handwriting OCR - Recognize handwritten text (limited accuracy)

🔴 This server has NO authentication by default!

Do NOT expose directly to the internet!

For production deployment:

• ✅ Use reverse proxy (Nginx/Caddy) with HTTPS
• ✅ Add API key authentication
• ✅ Configure firewall rules
• ✅ Enable rate limiting
• ✅ Set up monitoring

• ❌ CPU-only (no GPU acceleration in this version)
• ❌ Single request processing (no parallel requests)
• ❌ No built-in authentication
• ❌ Model files (~2GB) not included in repository

# Manual download
mkdir -p models && cd models
git lfs install
git clone https://www.modelscope.cn/PaddlePaddle/PaddleOCR-VL.git PaddleOCR-VL-0.9B

• Reduce image size
• Lower max_tokens
• Restart service: ./quick_start.sh restart

# Check port usage
lsof -i :7777

# View logs
./quick_start.sh attach

paddleocr-vl-cpu/
├── paddleocr_api.py           # Main API server
├── requirements_frozen.txt    # Locked dependencies
├── quick_init_setup.sh        # Setup script
├── quick_start.sh             # Service management
├── python_version.txt         # Python version lock
└── models/                    # Model files (gitignored)

• PaddleOCR Team - For the excellent PaddleOCR-VL model
• @hjdhnx - For inspiration from this discussion
• FastAPI - Modern web framework
• Transformers - Hugging Face's ML library

MIT License - see LICENSE file for details.

Note: PaddleOCR-VL model has its own license. Check the model repository for details.

• 🐛 Bug Reports: GitHub Issues
• 💬 Discussions: GitHub Discussions
• ⭐ Star this project if you find it helpful!

打破 GPU 限制！本项目让你能在 CPU 上运行 PaddleOCR-VL（飞桨团队优秀的 OCR 模型），并提供 OpenAI 兼容的 API 接口。

• 🖥️ CPU 友好 - 针对 CPU 推理优化，自动内存管理
• 🔌 OpenAI 兼容 - 可直接替换 OpenAI 视觉 API
• 🚀 快速部署 - 一条命令完成安装和启动
• ⚡ 流式支持 - 实时流式响应
• 💾 自动清理 - 防止长时间运行的内存泄漏
• 🖼️ 多格式支持 - 支持 JPEG、PNG、BMP、Base64、URL

• Python 3.12.3
• 16GB+ 内存（CPU 推理推荐配置）
• Ubuntu/Linux（Windows 用户请使用 WSL）

# 克隆仓库
git clone https://github.com/Think-Core/paddleocr-vl-cpu.git
cd paddleocr-vl-cpu

# 一键安装（自动下载模型 约2GB）
chmod +x quick_init_setup.sh
./quick_init_setup.sh

脚本会自动完成：

1. ✅ 创建 Python 虚拟环境
2. ✅ 安装所有依赖
3. ✅ 下载 PaddleOCR-VL-0.9B 模型（约 2GB）
4. ✅ 验证安装

# 启动服务器
./quick_start.sh start

# 检查状态
./quick_start.sh status

# 查看日志
./quick_start.sh attach  # 按 Ctrl+B 然后 D 退出

# 停止服务
./quick_start.sh stop

服务运行在 http://localhost:7777

import requests
import base64

# 读取并编码图片
with open("document.jpg", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode()

# OCR 请求
response = requests.post(
    "http://localhost:7777/v1/chat/completions",
    json={
        "model": "paddleocr-vl",
        "messages": [{
            "role": "user",
            "content": [
                {"type": "text", "text": "识别这张图片中的所有文字"},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": f"data:image/jpeg;base64,{image_b64}"
                    }
                }
            ]
        }],
        "max_tokens": 2048
    }
)

print(response.json()["choices"][0]["message"]["content"])

import json

response = requests.post(
    "http://localhost:7777/v1/chat/completions",
    json={
        "model": "paddleocr-vl",
        "messages": [...],  # 同上
        "stream": True
    },
    stream=True
)

for line in response.iter_lines():
    if line:
        line = line.decode('utf-8')
        if line.startswith('data: '):
            data = line[6:]
            if data != '[DONE]':
                chunk = json.loads(data)
                content = chunk["choices"][0]["delta"].get("content", "")
                print(content, end='', flush=True)

curl -X POST http://localhost:7777/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "paddleocr-vl",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "识别文字"},
        {
          "type": "image_url",
          "image_url": {"url": "data:image/jpeg;base64,{image_b64}"}
        }
      ]
    }]
  }'

• 📄 文档数字化 - 将扫描文档转换为文本
• 🧾 发票处理 - 自动提取发票信息
• 📋 表单识别 - 解析结构化表单和表格
• 🔍 证件识别 - 提取身份证、证书等文字
• 📚 书籍扫描 - 数字化纸质书籍和文章
• 📝 手写识别 - 识别手写文字（准确度有限）

🔴 服务器默认无身份验证！

切勿直接暴露到公网！

生产部署建议：

• ✅ 使用反向代理（Nginx/Caddy）配置 HTTPS
• ✅ 添加 API 密钥验证
• ✅ 配置防火墙规则
• ✅ 启用速率限制
• ✅ 设置监控告警

• ❌ 仅支持 CPU（本版本不含 GPU 加速）
• ❌ 单次请求处理（不支持并发请求）
• ❌ 无内置身份验证
• ❌ 模型文件（约 2GB）不包含在仓库中

# 手动下载
mkdir -p models && cd models
git lfs install
git clone https://www.modelscope.cn/PaddlePaddle/PaddleOCR-VL.git PaddleOCR-VL-0.9B

• 压缩图片尺寸
• 降低 max_tokens
• 重启服务：./quick_start.sh restart

# 检查端口占用
lsof -i :7777

# 查看日志
./quick_start.sh attach

paddleocr-vl-cpu/
├── paddleocr_api.py           # API 服务器主程序
├── requirements_frozen.txt    # 锁定的依赖版本
├── quick_init_setup.sh        # 安装脚本
├── quick_start.sh             # 服务管理脚本
├── python_version.txt         # Python 版本锁定
└── models/                    # 模型文件（已忽略）

• PaddleOCR 团队 - 提供出色的 PaddleOCR-VL 模型
• @hjdhnx - 来自 这个讨论 的启发
• FastAPI - 现代化 Web 框架
• Transformers - Hugging Face 的 ML 库

MIT 许可证 - 详见 LICENSE 文件。

注意： PaddleOCR-VL 模型有自己的许可证，请查看 模型仓库 了解详情。

• 🐛 错误报告：GitHub Issues
• 💬 问题讨论：GitHub Discussions
• ⭐ 觉得有用请点星！