Views
No views yet
ds4sd/docling-models
and ds4sd/docling-layout-heron-101.| File | Source | Description |
|---|---|---|
layout.onnx | ds4sd/docling-layout-heron-101 | RT-DETR layout detector (DocLayNet, 11 classes) |
tableformer_encoder.onnx | ds4sd/docling-models (tableformer/accurate) | TableFormer encoder, image → memory |
tableformer_decoder.onnx | ds4sd/docling-models (tableformer/accurate) | TableFormer autoregressive decoder, (tokens, memory) → (next_token, bbox) |
tableformer_vocab.json | derived | OTSL vocabulary for the decoder |
1# Gemfile
2gem "rb_docling", "~> 0.1"1# Scarica i modelli ONNX di Docling
2bundle exec rake models:fetch1require "rb_docling"
2
3tree = RbDocling.parse("doc.pdf",
4 layout: :onnx, table: :onnx,
5 models_dir: "./models")
6puts tree.to_md| Source PyTorch version | torch 2.x |
| ONNX opset | 17 |
| Quantization | none (FP32) |
| Conversion scripts | tools/ in the rb_docling repo |
1git clone https://github.com/scinoky/rb_docling
2cd rb_docling
3pip install -r tools/requirements.txt
4python tools/export_layout.py
5python tools/export_tableformer.py --variant accurate --mode split1@article{Docling,
2 title = {Docling Technical Report},
3 author = {Auer, Christoph and Lysak, Maksym and others (IBM Research)},
4 journal = {arXiv preprint},
5 year = {2024}
6}