Docling¶

Docling is a great Python library for parsing documents of diverse formats. It features a plugins system that allows extending its capabilities with custom models.

Using Docling is the recommended way to integrate our pipelines and models in a complete pipeline, with native text extraction from PDFs, OCR, picture annotation, serialization to formats such as Markdown, and hierarchical chunking.

CLI¶

Starting from Docling v2.120.0, table structure and layout plugins can be used directly with docling CLI.

To check the available external plugins, one can run:

docling --show-external-plugins

If cells2table is properly installed, it should show some plugins:

           Available layout engines
┏━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┓
┃          Name ┃ Plugin        ┃ Package     ┃
┡━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━┩
│ ppdoclayoutv3 │ ppdoclayoutv3 │ cells2table │
└───────────────┴───────────────┴─────────────┘
          Available table engines
┏━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━━━┓
┃        Name ┃ Plugin      ┃ Package     ┃
┡━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━━━┩
│ cells2table │ cells2table │ cells2table │
└─────────────┴─────────────┴─────────────┘

Those plugins can then be used easily:

docling --allow-external-plugins --table-structure-engine=cells2table --layout-engine=ppdoclayoutv3 <source>

Python¶

Using from Python is also easy:

from cells2table.docling import (
    CustomDoclingLayoutOptions,
    CustomDoclingTableStructureOptions,
)

pipeline_options = PdfPipelineOptions(
    allow_external_plugins=True,
    layout_options=CustomDoclingLayoutOptions(),
    table_structure_options=CustomDoclingTableStructureOptions(),
)

converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options),
        InputFormat.IMAGE: ImageFormatOption(pipeline_options=pipeline_options),
    }
)

result = converter.convert("path/to/document.pdf")
print(result.document.export_to_markdown())