Granite Docling 258M
Granite Docling 258M converts document images into structured DocTags text. Docling can then turn that representation into formats such as Markdown or HTML. The checkpoint is a document conversion component.
Model design and capabilities
The visual encoder is SigLIP2-base-patch16-512. The model is designed for document conversion rather than general image conversation. Its generated structure is consumed by DoclingDocument. The checkpoint does not replace every part of the Docling processing library.Granite Docling 258M model card
Practical use
Practical guidance. Use Docling when reading order and document structure must survive conversion. Inspect headings, lists, tables and equations after export, not only the recognized text. Keep the original page image available so a reviewer can resolve conversion mistakes.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-docling-258M |
| License | Apache 2.0Granite Docling 258M model card |
| Parameters | 258MGranite Docling 258M model card |
| Architecture | Idefics3 design with a SigLIP2 base encoder and Granite 165M language model.Granite Docling 258M model card |
Related models
- Granite Vision 4.1 4B
IBM · Granite Vision
An approximately 4B image-and-text model for chart, table and document understanding.
Launch date not confirmedAPI · Downloadable · Provider product