DeepSeek OCR Ollama

This is a simple script that uses the Ollama API to get the Markdown text from a PDF or image file using the DeepSeek-OCR model

Usage

Install the Requirements

To install the necessary requirements, run the following command:

pip install deepseek-ocr-ollama

To be used, Ollama is required and the deepseek-ocr model must be installed

ollama pull deepseek-ocr

Typical Usage

deepseek-ocr-ollama paper.pdf
deepseek-ocr-ollama paper.pdf --dpi 200
deepseek-ocr-ollama paper.pdf -o revision
deepseek-ocr-ollama paper.pdf -e
deepseek-ocr-ollama paper.pdf -m FULL
deepseek-ocr-ollama page74.jpg -e
deepseek-ocr-ollama OLLAMA_HOST=http://gauss:11434 receipt.pdf -e
deepseek-ocr-ollama -j paper.json
deepseek-ocr-ollama -j paper.json -m TEXT_NO_PAGES -n

Arguments

Modes

Given the input file paper.pdf, the directory structure for each mode is shown below:

0 - FULL

Structure

paper
├── full
│   ├── image1.png
│   ├── image2.png
│   ├── image3.png
│   └── paper.md
├── page_0
│   ├── image1.png
│   └── paper.md
├── page_1
│   ├── image2.png
│   └── paper.md
└── page_2
    ├── image3.png
    └── paper.md

1 - FULL_ALT

Structure

paper
├── image1.png
├── image2.png
├── image3.png
├── paper.md
├── page_0
│   ├── image1.png
│   └── paper.md
├── page_1
│   ├── image2.png
│   └── paper.md
└── page_2
    ├── image3.png
    └── paper.md

2 - FULL_NO_DIR

Structure

paper
├── image1.png
├── image2.png
├── image3.png
├── paper.md
├── paper0.md
├── paper1.md
└── paper2.md

3 - FULL_NO_PAGES default

Structure

paper
├── image1.png
├── image2.png
├── image3.png
└── paper.md

4 - TEXT

Structure

paper
├── paper.md
├── paper0.md
├── paper1.md
└── paper2.md

5 - TEXT_NO_PAGES

Structure

paper
└── paper.md

By default, the JSON response from the DeepSeek-OCR model is saved in the output directory. To disable JSON output, use the -n or --no-json argument. To experiment with a different mode without using additional calls, reuse an existing JSON response instead of the original input file

Ollama's Environment Variables

The Ollama server can be modified using the environment variables available from the Python API:

• OLLAMA_HOST : Ollama server host
• OLLAMA_API_KEY : Used as Bearer authorization token

To avoid using -e to load the .env file, you can create one at $HOME/.deepseek_ocr_ollama.env (where $HOME is your home directory). It will then be automatically loaded at the start of the script

For example, for an user called vavilov, the path would look like this:

• Linux

  /home/vavilov/.deepseek_ocr_ollama.env  
  

• macOS

  /Users/vavilov/.deepseek_ocr_ollama.env  
  

• Windows

  C:\Users\vavilov\.deepseek_ocr_ollama.env  
  

and the content will be something like this:

OLLAMA_HOST=http://gauss:11434