rb_docling-onnx

rb-docling-onnx

Pre-converted ONNX weights of Docling models for use with rb_docling (Ruby port).

These are third-party conversions of the official Docling PyTorch checkpoints. For the canonical PyTorch weights see ds4sd/docling-models and ds4sd/docling-layout-heron-101.

Files

Usage from Ruby

# Gemfile
gem "rb_docling", "~> 0.1"

# Scarica i modelli ONNX di Docling
bundle exec rake models:fetch

require "rb_docling"

tree = RbDocling.parse("doc.pdf",
                       layout: :onnx, table: :onnx,
                       models_dir: "./models")
puts tree.to_md

Conversion details

To re-build from the official PyTorch weights:

git clone https://github.com/scinoky/rb_docling
cd rb_docling
pip install -r tools/requirements.txt
python tools/export_layout.py
python tools/export_tableformer.py --variant accurate --mode split

License

Apache-2.0, inherited from the upstream Docling models.

Citation

If you use these conversions, please cite the original Docling work:

@article{Docling,
  title   = {Docling Technical Report},
  author  = {Auer, Christoph and Lysak, Maksym and others (IBM Research)},
  journal = {arXiv preprint},
  year    = {2024}
}