Docling Layout Heron (GGUF for LM-Kit)

A GGUF conversion of docling-project/docling-layout-heron, an RT-DETRv2 document layout detector. It finds the regions of a page with their bounds and confidence, across 17 classes: caption, footnote, formula, list item, page footer, page header, picture, section header, table, text, title, document index, code, checkbox (selected and unselected), form and key-value region.

This file runs on LM-Kit.NET's graph runner (architecture lmkit-layout-rtdetr), on the CPU or a GPU, in full F32 precision. It is not a llama.cpp language model.

File	Precision	Size	SHA-256
`docling-layout-heron.gguf`	F32	168,456,096 bytes	`4c3ba02bb6697e91e738202c7f040c7535b51829ac93a261062c6618e3fd7ba8`

Input: 640 x 640 RGB. Parameters: 42.1M. Converter revision 1, from the upstream weights with SHA-256 59c81a3a2923042d85034ffc487f8f47e4854117e879aef89b2b9f728fb4922a.

Usage with LM-Kit.NET

using LMKit.Model;

LM layoutModel = LM.LoadFromModelID("docling-layout-heron");
// layoutModel.HasLayoutDetection == true

License and attribution

Apache-2.0, as the upstream model. Credit goes to the Docling project. If you use this model, please cite the upstream works:

Livathinos et al., "Advanced Layout Analysis Models for Docling", 2025, arXiv:2509.11720.
Deep Search Team, "Docling Technical Report", 2024, arXiv:2408.09869.