DS4SD / docling

Get your documents ready for gen AI
https://ds4sd.github.io/docling
MIT License
11.01k stars 536 forks source link

Docling crashes when using EasyOCR on Windows 11 #318

Closed cau-git closed 17 hours ago

cau-git commented 1 week ago

Bug

On a Windows 11 installation using ARM64 CPU (UTM virtual machine on macOS host), docling silently crashes without generating output when a document is converted with OCR enabled (using default EasyOCR).

❓ Everyone in the community, can you please help us reproducing this on a Windows x86 (64bit) machine and report if this is broken? 🙏 ⚠️ Note: This is observed with the native Windows docling package, not in WSL.

Steps to reproduce

  1. Install windows-native Python (3.12) through Microsoft Store
  2. Set up a fresh virtualenv on a Windows system with python 3.12 and activate
    python -m venv venv
    venv\Scripts\activate.bat
  3. Install docling through pip install docling
  4. Run docling on a command prompt with: docling https://arxiv.org/pdf/2206.01062.pdf, observe silent crash and no output.
  5. Couter-check: Running docling https://arxiv.org/pdf/2206.01062.pdf --no-ocr should work fine.

Docling version

Docling version: 2.5.0
Docling Core version: 2.3.2
Docling IBM Models version: 2.0.3
Docling Parse version: 2.0.3

Python version

Python 3.12.7
maxmnemonic commented 1 week ago

I checked on Windows 11 Pro 64bit x86 bare metal machine. Both commands (with OCR and without OCR) worked fine, and produced MD files

jokus-pokus commented 1 week ago

Hey @cau-git have you checked if your machine runs out of resources maybe? I had issues with vision models on an arm machine as well and never found the issue.

cau-git commented 17 hours ago

This seems to be non-reproducible on any other environments than a Windows 11 ARM VM on UTM. Closing.