Skip to content

Docling v2

What's new

Docling v2 introduces several new features:

  • Understands and converts PDF, MS Word, MS Powerpoint, HTML and several image formats
  • Produces a new, universal document representation which can encapsulate document hierarchy
  • Comes with a fresh new API and CLI

Changes in Docling v2

CLI

We updated the command line syntax of Docling v2 to support many formats. Examples are seen below. 

# Convert a single file to Markdown (default)
docling myfile.pdf

# Convert a single file to Markdown and JSON, without OCR
docling myfile.pdf --to json --to md --no-ocr

# Convert PDF files in input directory to Markdown (default)
docling ./input/dir --from pdf

# Convert PDF and Word files in input directory to Markdown and JSON
docling ./input/dir --from pdf --from docx --to md --to json --output ./scratch

# Convert all supported files in input directory to Markdown, but abort on first error
docling ./input/dir --output ./scratch --abort-on-error

Notable changes from Docling v1:

  • The standalone switches for different export formats are removed, and replaced with --from and --to arguments, to define input and output formats respectively.
  • The new --abort-on-error will abort any batch conversion as soon an error is encountered
  • The --backend option for PDFs was removed

Setting up a DocumentConverter

To accomodate many input formats, we changed the way you need to set up your DocumentConverter object. You can now define a list of allowed formats on the DocumentConverter initialization, and specify custom options per-format if desired. By default, all supported formats are allowed. If you don't provide format_options, defaults will be used for all allowed_formats.

Format options can include the pipeline class to use, the options to provide to the pipeline, and the document backend. They are provided as format-specific types, such as PdfFormatOption or WordFormatOption, as seen below.

from docling.document_converter import DocumentConverter
from docling.datamodel.base_models import InputFormat
from docling.document_converter import (
    DocumentConverter,
    PdfFormatOption,
    WordFormatOption,
)
from docling.pipeline.simple_pipeline import SimplePipeline
from docling.pipeline.standard_pdf_pipeline import StandardPdfPipeline
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend

## Default initialization still works as before:
# doc_converter = DocumentConverter()


# previous `PipelineOptions` is now `PdfPipelineOptions`
pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = False
pipeline_options.do_table_structure = True
#...

## Custom options are now defined per format.
doc_converter = (
    DocumentConverter(  # all of the below is optional, has internal defaults.
        allowed_formats=[
            InputFormat.PDF,
            InputFormat.IMAGE,
            InputFormat.DOCX,
            InputFormat.HTML,
            InputFormat.PPTX,
        ],  # whitelist formats, non-matching files are ignored.
        format_options={
            InputFormat.PDF: PdfFormatOption(
                pipeline_options=pipeline_options, # pipeline options go here.
                backend=PyPdfiumDocumentBackend # optional: pick an alternative backend
            ),
            InputFormat.DOCX: WordFormatOption(
                pipeline_cls=SimplePipeline # default for office formats and HTML
            ),
        },
    )
)

Note: If you work only with defaults, all remains the same as in Docling v1.

More options are shown in the following example units:

Converting documents

We have simplified the way you can feed input to the DocumentConverter and renamed the conversion methods for better semantics. You can now call the conversion directly with a single file, or a list of input files, or DocumentStream objects, without constructing a DocumentConversionInput object first.

  • DocumentConverter.convert now converts a single file input (previously DocumentConverter.convert_single).
  • DocumentConverter.convert_all now converts many files at once (previously DocumentConverter.convert).

...
from docling.datamodel.document import ConversionResult
## Convert a single file (from URL or local path)
conv_result: ConversionResult = doc_converter.convert("https://arxiv.org/pdf/2408.09869") # previously `convert_single`

## Convert several files at once:

input_files = [
    "tests/data/wiki_duck.html",
    "tests/data/word_sample.docx",
    "tests/data/lorem_ipsum.docx",
    "tests/data/powerpoint_sample.pptx",
    "tests/data/2305.03393v1-pg9-img.png",
    "tests/data/2206.01062.pdf",
]

# Directly pass list of files or streams to `convert_all`
conv_results_iter = doc_converter.convert_all(input_files) # previously `convert`
Through the raises_on_error argument, you can also control if the conversion should raise exceptions when first encountering a problem, or resiliently convert all files first and reflect errors in each file's conversion status. By default, any error is immediately raised and the conversion aborts (previously, exceptions were swallowed).

...
conv_results_iter = doc_converter.convert_all(input_files, raises_on_error=False) # previously `convert`

Access document structures

We have simplified how you can access and export the converted document data, too. Our universal document representation is now available in conversion results as a DoclingDocument object. DoclingDocument provides a neat set of APIs to construct, iterate and export content in the document, as shown below.

conv_result: ConversionResult = doc_converter.convert("https://arxiv.org/pdf/2408.09869") # previously `convert_single`

## Inspect the converted document:
conv_result.document.print_element_tree()

## Iterate the elements in reading order, including hierachy level:
for item, level in conv_result.document.iterate_items:
    if isinstance(item, TextItem):
        print(item.text)
    elif isinstance(item, TableItem):
        table_df: pd.DataFrame = item.export_to_dataframe()
        print(table_df.to_markdown())
    elif ...:
        #...

Note: While it is deprecated, you can still work with the Docling v1 document representation, it is available as: 

conv_result.legacy_document # provides the representation in previous ExportedCCSDocument type

Export into JSON, Markdown, Doctags

Note: All render_... methods in ConversionResult have been removed in Docling v2, and are now available on DoclingDocument as:

  • DoclingDocument.export_to_dict
  • DoclingDocument.export_to_markdown
  • DoclingDocument.export_to_document_tokens
conv_result: ConversionResult = doc_converter.convert("https://arxiv.org/pdf/2408.09869") # previously `convert_single`

## Export to desired format:
print(json.dumps(conv_res.document.export_to_dict()))
print(conv_res.document.export_to_markdown())
print(conv_res.document.export_to_document_tokens())

Note: While it is deprecated, you can still export Docling v1 JSON format. This is available through the same methods as on the DoclingDocument type: 

## Export legacy document representation to desired format, for v1 compatibility:
print(json.dumps(conv_res.legacy_document.export_to_dict()))
print(conv_res.legacy_document.export_to_markdown())
print(conv_res.legacy_document.export_to_document_tokens())

Reload a DoclingDocument stored as JSON

You can save and reload a DoclingDocument to disk in JSON format using the following codes:

# Save to disk:
doc: DoclingDocument = conv_res.document # produced from conversion result...

with Path("./doc.json").open("w") as fp:
    fp.write(json.dumps(doc.export_to_dict())) # use `export_to_dict` to ensure consistency

# Load from disk:
with Path("./doc.json").open("r") as fp:
    doc_dict = json.loads(fp.read())
    doc = DoclingDocument.model_validate(doc_dict) # use standard pydantic API to populate doc

Chunking

Docling v2 defines new base classes for chunking:

  • BaseMeta for chunk metadata
  • BaseChunk containing the chunk text and metadata, and
  • BaseChunker for chunkers, producing chunks out of a DoclingDocument.

Additionally, it provides an updated HierarchicalChunker implementation, which leverages the new DoclingDocument and provides a new, richer chunk output format, including:

  • the respective doc items for grounding
  • any applicable headings for context
  • any applicable captions for context

For an example, check out Chunking usage.