markmagic

convert files into markdown

Supported file types (and processing engine):

• docx (python-docx)
• eml (email)
• excel (python-calamine)
• pdf (pypdf / vision-language-models)
• pptx (python-pptx)

Getting started

from pathlib import Path
from markmagic import convert_any

with Path("tests/data/docx/msft_pr.docx").open("rb") as f:
    convert_any(filename="msft_pr.docx", file=f)

If you're interested in using vision language models to ocr a pdf

Create a .env file in the root directory

API_KEY="REPLACE"

from pathlib import Path
from markmagic import Settings, convert_any

settings = Settings(use_vlm=True)
with Path("tests/data/pdf/msft_ar.pdf").open("rb") as f:
    convert_any(filename="msft_ar.pdf", file=f, settings=settings)

Design / Limitations

• markmagic only looks at the file extension to decide how to convert your files
• markmagic uses python-docx so cannot extract text from shapes / images (consider using python-mammoth + markdownify)

Goals / Motivation

• OCR is just neural nets so why not just use vision language models for ocr?
• OCRBenchmark https://github.com/open-compass/VLMEvalKit?tab=readme-ov-file