docling-models is an open-weight model from Docling, released under cdla-permissive-2.0. Its published files total 358.2 MB. It draws 1.5M downloads a month.
This page contains models that power the PDF document converion package docling. The layout model will take an image from a page and apply RT-DETR model in order to find different layout components.
SAVRN's Notes on docling-models
Not one model but the bundle behind the docling PDF converter: an RT-DETR layout model that detects eleven component types — Caption, Footnote, Formula, List-item, Page-footer, Page-header, Picture, Section-header, Table, Text, Title — plus a TableFormer for table structure. Together they total 358.2 MB across two safetensors files, small by hub standards, under cdla-permissive-2.0. First published 2024-07-02 and still updated — last revision dated 2025-12-03 — with 1.5M downloads a month riding along with the library.
Deployment is library-shaped, not server-shaped: install docling and these weights come along, so there is no Runs On table and no per-token host prices on our Index. The card's DocLayNet reference puts human layout accuracy at 84–89 against high-70s for the best detectors — a useful ceiling when judging extraction quality. If your workload is document conversion, self-hosting this bundle beats a managed vision API on cost per page; if you only need one component, the granularity is here to pull it apart.
Model Card
By Docling, published under cdla-permissive-2.0, revision 2199320848bb.
This page contains models that power the PDF document converion package docling.
Layout Model
The layout model will take an image from a page and apply RT-DETR model in order to find different layout components. It currently detects the labels: Caption, Footnote, Formula, List-item, Page-footer, Page-header, Picture, Section-header, Table, Text, Title. As a reference (from the DocLayNet-paper), this is the performance of standard object detection methods on the DocLayNet dataset compared to human evaluation,
| human | MRCNN | MRCNN | FRCNN | YOLO | |
|---|---|---|---|---|---|
| human | R50 | R101 | R101 | v5x6 | |
| Caption | 84-89 | 68.4 | 71.5 | 70.1 | 77.7 |
| Footnote | 83-91 | 70.9 | 71.8 | 73.7 | 77.2 |
| Formula | 83-85 | 60.1 | 63.4 | 63.5 | 66.2 |
| List-item | 87-88 | 81.2 | 80.8 | 81.0 | 86.2 |
| Page-footer | 93-94 | 61.6 | 59.3 | 58.9 | 61.1 |
| Page-header | 85-89 | 71.9 | 70.0 | 72.0 | 67.9 |
| Picture | 69-71 | 71.7 | 72.7 | 72.0 | 77.1 |
| Section-header | 83-84 | 67.6 | 69.3 | 68.4 | 74.6 |
| Table | 77-81 | 82.2 | 82.9 | 82.2 | 86.3 |
| Text | 84-86 | 84.6 | 85.8 | 85.4 | 88.1 |
| Title | 60-72 | 76.7 | 80.4 | 79.9 | 82.7 |
| All | 82-83 | 72.4 | 73.5 | 73.4 | 76.8 |
TableFormer
Identity and Version
- Repository
- docling-project/docling-models
- Publisher
- Docling
- Task
- Not stated by the source
- Modality
- Other
- Library
- transformers
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 2199320848bb9a8a519d22e4b528185a4f9a6f64
- First published
- 2024-07-02
- Last updated
- 2025-12-03
Files and Weights
8 files, 358.2 MB in total. The weights are 2 files totalling 358.2 MB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model_artifacts/tableformer/accurate/tableformer_accurate.safetensors | Weights | 212.8 MB | 2a7d6c924b3c |
| model_artifacts/tableformer/fast/tableformer_fast.safetensors | Weights | 145.5 MB | 3119563aab5a |
| config.json | Configuration | 41 B | — |
| model_artifacts/tableformer/accurate/tm_config.json | Configuration | 7.1 KB | — |
| model_artifacts/tableformer/fast/tm_config.json | Configuration | 7.1 KB | — |
| README.md | Documentation | 3.4 KB | — |
| .gitattributes | Repository | 1.9 KB | — |
| .gitignore | Repository | 5.2 KB | — |
License and Download
- License
- cdla-permissive-2.0
- Access
- Open weights, no gate
- Download size
- 358.2 MB
Released by Docling through its official repository on Hugging Face.
Built From
- Described by arXiv:2206.01062
- Described by arXiv:2408.09869
Evaluations
Each result is shown as reported, with the conditions its reporter stated. None is a SAVRN measurement. A comparison lines two results up only when their configuration, unit and setup are all stated and identical.
| Benchmark | Conditions | Result | Reported by | Revision | Date |
|---|---|---|---|---|---|
| llamaindex/ParseBench | Task chartMetric chartSetup Pipeline name: docling_parseComparison conditions not established | 52.8 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-03-31 |
| llamaindex/ParseBench | Task layoutMetric layoutSetup Pipeline name: docling_parseComparison conditions not established | 66.1 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-09 |
| llamaindex/ParseBench | Task meanMetric meanSetup Pipeline name: docling_parseComparison conditions not established | 50.6 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-09 |
| llamaindex/ParseBench | Task tableMetric tableSetup Pipeline name: docling_parseComparison conditions not established | 66.4 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-08 |
| llamaindex/ParseBench | Task text_contentMetric text_contentSetup Pipeline name: docling_parseComparison conditions not established | 66.9 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-03-31 |
| llamaindex/ParseBench | Task text_formattingMetric text_formattingSetup Pipeline name: docling_parseComparison conditions not established | 1 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-03-31 |
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 358.2 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About docling-models
What license is docling-models released under?
cdla-permissive-2.0, as its publisher declares it. Read the license text before commercial use.