Pipeline options
Pipeline options allow to customize the execution of the models during the conversion pipeline.
This includes options for the OCR engines, the table model as well as enrichment options which
can be enabled with do_xyz = True.
This is an automatic generated API reference of the all the pipeline options available in Docling.
Classes:
-
AsrPipelineOptions– -
BaseLayoutOptions–Base options for layout models.
-
BaseOptions–Base class for options.
-
BaseTableStructureOptions–Base options for table structure models.
-
ConvertPipelineOptions–Base convert pipeline options.
-
EasyOcrOptions–Options for the EasyOCR engine.
-
LayoutOptions–Options for layout processing.
-
OcrAutoOptions–Options for pick OCR engine automatically.
-
OcrEngine–Enum of valid OCR engines.
-
OcrMacOptions–Options for the Mac OCR engine.
-
OcrOptions–OCR options.
-
PaginatedPipelineOptions– -
PdfBackend–Enum of valid PDF backends.
-
PdfPipelineOptions–Options for the PDF pipeline.
-
PictureDescriptionApiOptions– -
PictureDescriptionBaseOptions– -
PictureDescriptionVlmOptions– -
PipelineOptions–Base pipeline options.
-
ProcessingPipeline– -
RapidOcrOptions–Options for the RapidOCR engine.
-
TableFormerMode–Modes for the TableFormer model.
-
TableStructureOptions–Options for the table structure.
-
TesseractCliOcrOptions–Options for the TesseractCli engine.
-
TesseractOcrOptions–Options for the Tesseract engine.
-
ThreadedPdfPipelineOptions–Pipeline options for the threaded PDF pipeline with batching and backpressure control
-
VlmExtractionPipelineOptions–Options for extraction pipeline.
-
VlmPipelineOptions–
Attributes:
module-attribute
granite_picture_description = PictureDescriptionVlmOptions(repo_id='ibm-granite/granite-vision-3.3-2b', prompt='What is shown in this image?')
module-attribute
smolvlm_picture_description = PictureDescriptionVlmOptions(repo_id='HuggingFaceTB/SmolVLM-256M-Instruct')
pydantic-model
Bases: PipelineOptions
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"InlineAsrOptions": {
"properties": {
"kind": {
"const": "inline_model_options",
"default": "inline_model_options",
"title": "Kind",
"type": "string"
},
"repo_id": {
"title": "Repo Id",
"type": "string"
},
"verbose": {
"default": false,
"title": "Verbose",
"type": "boolean"
},
"timestamps": {
"default": true,
"title": "Timestamps",
"type": "boolean"
},
"temperature": {
"default": 0.0,
"title": "Temperature",
"type": "number"
},
"max_new_tokens": {
"default": 256,
"title": "Max New Tokens",
"type": "integer"
},
"max_time_chunk": {
"default": 30.0,
"title": "Max Time Chunk",
"type": "number"
},
"torch_dtype": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Torch Dtype"
},
"supported_devices": {
"default": [
"cpu",
"cuda",
"mps"
],
"items": {
"$ref": "#/$defs/AcceleratorDevice"
},
"title": "Supported Devices",
"type": "array"
}
},
"required": [
"repo_id"
],
"title": "InlineAsrOptions",
"type": "object"
}
},
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"asr_options": {
"$ref": "#/$defs/InlineAsrOptions",
"default": {
"kind": "inline_model_options",
"repo_id": "tiny",
"verbose": true,
"timestamps": true,
"temperature": 0.0,
"max_new_tokens": 256,
"max_time_chunk": 30.0,
"torch_dtype": null,
"supported_devices": [
"cpu",
"cuda"
],
"inference_framework": "whisper",
"language": "en",
"word_timestamps": true
}
}
},
"title": "AsrPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
asr_options(Union[InlineAsrOptions])
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
asr_options: Union[InlineAsrOptions] = WHISPER_TINY
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
class-attribute
kind: str
pydantic-model
Bases: BaseOptions
Base options for layout models.
Show JSON schema:
{
"description": "Base options for layout models.",
"properties": {
"keep_empty_clusters": {
"default": false,
"title": "Keep Empty Clusters",
"type": "boolean"
},
"skip_cell_assignment": {
"default": false,
"title": "Skip Cell Assignment",
"type": "boolean"
}
},
"title": "BaseLayoutOptions",
"type": "object"
}
Fields:
-
keep_empty_clusters(bool) -
skip_cell_assignment(bool)
pydantic-field
keep_empty_clusters: bool = False
class-attribute
kind: str
pydantic-field
skip_cell_assignment: bool = False
pydantic-model
Bases: BaseModel
Base class for options.
Show JSON schema:
{
"description": "Base class for options.",
"properties": {},
"title": "BaseOptions",
"type": "object"
}
class-attribute
kind: str
pydantic-model
Bases: BaseOptions
Base options for table structure models.
Show JSON schema:
{
"description": "Base options for table structure models.",
"properties": {},
"title": "BaseTableStructureOptions",
"type": "object"
}
class-attribute
kind: str
pydantic-model
Bases: PipelineOptions
Base convert pipeline options.
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"PictureDescriptionBaseOptions": {
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
}
},
"description": "Base convert pipeline options.",
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"do_picture_classification": {
"default": false,
"title": "Do Picture Classification",
"type": "boolean"
},
"do_picture_description": {
"default": false,
"title": "Do Picture Description",
"type": "boolean"
},
"picture_description_options": {
"$ref": "#/$defs/PictureDescriptionBaseOptions",
"default": {
"batch_size": 8,
"scale": 2.0,
"picture_area_threshold": 0.05,
"repo_id": "HuggingFaceTB/SmolVLM-256M-Instruct",
"prompt": "Describe this image in a few sentences.",
"generation_config": {
"do_sample": false,
"max_new_tokens": 200
}
}
}
},
"title": "ConvertPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
do_picture_classification(bool) -
do_picture_description(bool) -
picture_description_options(PictureDescriptionBaseOptions)
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
do_picture_classification: bool = False
pydantic-field
do_picture_description: bool = False
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
class-attribute
kind: str
pydantic-field
picture_description_options: PictureDescriptionBaseOptions = smolvlm_picture_description
pydantic-model
Bases: OcrOptions
Options for the EasyOCR engine.
Show JSON schema:
{
"additionalProperties": false,
"description": "Options for the EasyOCR engine.",
"properties": {
"lang": {
"default": [
"fr",
"de",
"es",
"en"
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
},
"use_gpu": {
"anyOf": [
{
"type": "boolean"
},
{
"type": "null"
}
],
"default": null,
"title": "Use Gpu"
},
"confidence_threshold": {
"default": 0.5,
"title": "Confidence Threshold",
"type": "number"
},
"model_storage_directory": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Model Storage Directory"
},
"recog_network": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": "standard",
"title": "Recog Network"
},
"download_enabled": {
"default": true,
"title": "Download Enabled",
"type": "boolean"
},
"suppress_mps_warnings": {
"default": true,
"title": "Suppress Mps Warnings",
"type": "boolean"
}
},
"title": "EasyOcrOptions",
"type": "object"
}
Config:
extra:forbidprotected_namespaces:()
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(List[str]) -
use_gpu(Optional[bool]) -
confidence_threshold(float) -
model_storage_directory(Optional[str]) -
recog_network(Optional[str]) -
download_enabled(bool) -
suppress_mps_warnings(bool)
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
confidence_threshold: float = 0.5
pydantic-field
download_enabled: bool = True
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: Literal['easyocr'] = 'easyocr'
pydantic-field
lang: List[str] = ['fr', 'de', 'es', 'en']
class-attribute
instance-attribute
model_config = ConfigDict(extra='forbid', protected_namespaces=())
pydantic-field
model_storage_directory: Optional[str] = None
pydantic-field
recog_network: Optional[str] = 'standard'
pydantic-field
suppress_mps_warnings: bool = True
pydantic-field
use_gpu: Optional[bool] = None
pydantic-model
Bases: BaseLayoutOptions
Options for layout processing.
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"LayoutModelConfig": {
"properties": {
"name": {
"title": "Name",
"type": "string"
},
"repo_id": {
"title": "Repo Id",
"type": "string"
},
"revision": {
"title": "Revision",
"type": "string"
},
"model_path": {
"title": "Model Path",
"type": "string"
},
"supported_devices": {
"default": [
"cpu",
"cuda",
"mps"
],
"items": {
"$ref": "#/$defs/AcceleratorDevice"
},
"title": "Supported Devices",
"type": "array"
}
},
"required": [
"name",
"repo_id",
"revision",
"model_path"
],
"title": "LayoutModelConfig",
"type": "object"
}
},
"description": "Options for layout processing.",
"properties": {
"keep_empty_clusters": {
"default": false,
"title": "Keep Empty Clusters",
"type": "boolean"
},
"skip_cell_assignment": {
"default": false,
"title": "Skip Cell Assignment",
"type": "boolean"
},
"create_orphan_clusters": {
"default": true,
"title": "Create Orphan Clusters",
"type": "boolean"
},
"model_spec": {
"$ref": "#/$defs/LayoutModelConfig",
"default": {
"name": "docling_layout_heron",
"repo_id": "docling-project/docling-layout-heron",
"revision": "main",
"model_path": "",
"supported_devices": [
"cpu",
"cuda",
"mps"
]
}
}
},
"title": "LayoutOptions",
"type": "object"
}
Fields:
-
keep_empty_clusters(bool) -
skip_cell_assignment(bool) -
create_orphan_clusters(bool) -
model_spec(LayoutModelConfig)
pydantic-field
create_orphan_clusters: bool = True
pydantic-field
keep_empty_clusters: bool = False
class-attribute
kind: str = 'docling_layout_default'
pydantic-field
model_spec: LayoutModelConfig = DOCLING_LAYOUT_HERON
pydantic-field
skip_cell_assignment: bool = False
pydantic-model
Bases: OcrOptions
Options for pick OCR engine automatically.
Show JSON schema:
{
"description": "Options for pick OCR engine automatically.",
"properties": {
"lang": {
"default": [],
"description": "The automatic OCR engine will use the default values of the engine. Please specify the engine explicitly to change the language selection.",
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
}
},
"title": "OcrAutoOptions",
"type": "object"
}
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(Annotated[List[str], Field(description='The automatic OCR engine will use the default values of the engine. Please specify the engine explicitly to change the language selection.')])
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: Literal['auto'] = 'auto'
pydantic-field
lang: Annotated[List[str], Field(description='The automatic OCR engine will use the default values of the engine. Please specify the engine explicitly to change the language selection.')] = []
Bases: str, Enum
Enum of valid OCR engines.
Attributes:
class-attribute
instance-attribute
AUTO = 'auto'
class-attribute
instance-attribute
EASYOCR = 'easyocr'
class-attribute
instance-attribute
OCRMAC = 'ocrmac'
class-attribute
instance-attribute
RAPIDOCR = 'rapidocr'
class-attribute
instance-attribute
TESSERACT = 'tesseract'
class-attribute
instance-attribute
TESSERACT_CLI = 'tesseract_cli'
pydantic-model
Bases: OcrOptions
Options for the Mac OCR engine.
Show JSON schema:
{
"additionalProperties": false,
"description": "Options for the Mac OCR engine.",
"properties": {
"lang": {
"default": [
"fr-FR",
"de-DE",
"es-ES",
"en-US"
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
},
"recognition": {
"default": "accurate",
"title": "Recognition",
"type": "string"
},
"framework": {
"default": "vision",
"title": "Framework",
"type": "string"
}
},
"title": "OcrMacOptions",
"type": "object"
}
Config:
extra:forbid
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(List[str]) -
recognition(str) -
framework(str)
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
pydantic-field
framework: str = 'vision'
class-attribute
kind: Literal['ocrmac'] = 'ocrmac'
pydantic-field
lang: List[str] = ['fr-FR', 'de-DE', 'es-ES', 'en-US']
class-attribute
instance-attribute
model_config = ConfigDict(extra='forbid')
pydantic-field
recognition: str = 'accurate'
pydantic-model
Bases: BaseOptions
OCR options.
Show JSON schema:
{
"description": "OCR options.",
"properties": {
"lang": {
"description": "List of OCR languages to use. The format must match the values of the OCR engine of choice.",
"examples": [
[
"deu",
"eng"
]
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
}
},
"required": [
"lang"
],
"title": "OcrOptions",
"type": "object"
}
Fields:
-
lang(Annotated[List[str], Field(description='List of OCR languages to use. The format must match the values of the OCR engine of choice.', examples=[[deu, eng]])]) -
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])])
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: str
pydantic-field
lang: Annotated[List[str], Field(description='List of OCR languages to use. The format must match the values of the OCR engine of choice.', examples=[[deu, eng]])]
pydantic-model
Bases: ConvertPipelineOptions
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"PictureDescriptionBaseOptions": {
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
}
},
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"do_picture_classification": {
"default": false,
"title": "Do Picture Classification",
"type": "boolean"
},
"do_picture_description": {
"default": false,
"title": "Do Picture Description",
"type": "boolean"
},
"picture_description_options": {
"$ref": "#/$defs/PictureDescriptionBaseOptions",
"default": {
"batch_size": 8,
"scale": 2.0,
"picture_area_threshold": 0.05,
"repo_id": "HuggingFaceTB/SmolVLM-256M-Instruct",
"prompt": "Describe this image in a few sentences.",
"generation_config": {
"do_sample": false,
"max_new_tokens": 200
}
}
},
"images_scale": {
"default": 1.0,
"title": "Images Scale",
"type": "number"
},
"generate_page_images": {
"default": false,
"title": "Generate Page Images",
"type": "boolean"
},
"generate_picture_images": {
"default": false,
"title": "Generate Picture Images",
"type": "boolean"
}
},
"title": "PaginatedPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
do_picture_classification(bool) -
do_picture_description(bool) -
picture_description_options(PictureDescriptionBaseOptions) -
images_scale(float) -
generate_page_images(bool) -
generate_picture_images(bool)
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
do_picture_classification: bool = False
pydantic-field
do_picture_description: bool = False
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
pydantic-field
generate_page_images: bool = False
pydantic-field
generate_picture_images: bool = False
pydantic-field
images_scale: float = 1.0
class-attribute
kind: str
pydantic-field
picture_description_options: PictureDescriptionBaseOptions = smolvlm_picture_description
Bases: str, Enum
Enum of valid PDF backends.
Attributes:
-
DLPARSE_V1– -
DLPARSE_V2– -
DLPARSE_V4– -
PYPDFIUM2–
class-attribute
instance-attribute
DLPARSE_V1 = 'dlparse_v1'
class-attribute
instance-attribute
DLPARSE_V2 = 'dlparse_v2'
class-attribute
instance-attribute
DLPARSE_V4 = 'dlparse_v4'
class-attribute
instance-attribute
PYPDFIUM2 = 'pypdfium2'
pydantic-model
Bases: PaginatedPipelineOptions
Options for the PDF pipeline.
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"BaseLayoutOptions": {
"description": "Base options for layout models.",
"properties": {
"keep_empty_clusters": {
"default": false,
"title": "Keep Empty Clusters",
"type": "boolean"
},
"skip_cell_assignment": {
"default": false,
"title": "Skip Cell Assignment",
"type": "boolean"
}
},
"title": "BaseLayoutOptions",
"type": "object"
},
"BaseTableStructureOptions": {
"description": "Base options for table structure models.",
"properties": {},
"title": "BaseTableStructureOptions",
"type": "object"
},
"OcrOptions": {
"description": "OCR options.",
"properties": {
"lang": {
"description": "List of OCR languages to use. The format must match the values of the OCR engine of choice.",
"examples": [
[
"deu",
"eng"
]
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
}
},
"required": [
"lang"
],
"title": "OcrOptions",
"type": "object"
},
"PictureDescriptionBaseOptions": {
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
}
},
"description": "Options for the PDF pipeline.",
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"do_picture_classification": {
"default": false,
"title": "Do Picture Classification",
"type": "boolean"
},
"do_picture_description": {
"default": false,
"title": "Do Picture Description",
"type": "boolean"
},
"picture_description_options": {
"$ref": "#/$defs/PictureDescriptionBaseOptions",
"default": {
"batch_size": 8,
"scale": 2.0,
"picture_area_threshold": 0.05,
"repo_id": "HuggingFaceTB/SmolVLM-256M-Instruct",
"prompt": "Describe this image in a few sentences.",
"generation_config": {
"do_sample": false,
"max_new_tokens": 200
}
}
},
"images_scale": {
"default": 1.0,
"title": "Images Scale",
"type": "number"
},
"generate_page_images": {
"default": false,
"title": "Generate Page Images",
"type": "boolean"
},
"generate_picture_images": {
"default": false,
"title": "Generate Picture Images",
"type": "boolean"
},
"do_table_structure": {
"default": true,
"title": "Do Table Structure",
"type": "boolean"
},
"do_ocr": {
"default": true,
"title": "Do Ocr",
"type": "boolean"
},
"do_code_enrichment": {
"default": false,
"title": "Do Code Enrichment",
"type": "boolean"
},
"do_formula_enrichment": {
"default": false,
"title": "Do Formula Enrichment",
"type": "boolean"
},
"force_backend_text": {
"default": false,
"title": "Force Backend Text",
"type": "boolean"
},
"table_structure_options": {
"$ref": "#/$defs/BaseTableStructureOptions",
"default": {
"do_cell_matching": true,
"mode": "accurate"
}
},
"ocr_options": {
"$ref": "#/$defs/OcrOptions",
"default": {
"lang": [],
"force_full_page_ocr": false,
"bitmap_area_threshold": 0.05
}
},
"layout_options": {
"$ref": "#/$defs/BaseLayoutOptions",
"default": {
"keep_empty_clusters": false,
"skip_cell_assignment": false,
"create_orphan_clusters": true,
"model_spec": {
"model_path": "",
"name": "docling_layout_heron",
"repo_id": "docling-project/docling-layout-heron",
"revision": "main",
"supported_devices": [
"cpu",
"cuda",
"mps"
]
}
}
},
"generate_table_images": {
"default": false,
"deprecated": true,
"title": "Generate Table Images",
"type": "boolean"
},
"generate_parsed_pages": {
"default": false,
"title": "Generate Parsed Pages",
"type": "boolean"
},
"ocr_batch_size": {
"default": 4,
"title": "Ocr Batch Size",
"type": "integer"
},
"layout_batch_size": {
"default": 4,
"title": "Layout Batch Size",
"type": "integer"
},
"table_batch_size": {
"default": 4,
"title": "Table Batch Size",
"type": "integer"
},
"batch_polling_interval_seconds": {
"default": 0.5,
"title": "Batch Polling Interval Seconds",
"type": "number"
},
"queue_max_size": {
"default": 100,
"title": "Queue Max Size",
"type": "integer"
}
},
"title": "PdfPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
do_picture_classification(bool) -
do_picture_description(bool) -
picture_description_options(PictureDescriptionBaseOptions) -
do_table_structure(bool) -
do_ocr(bool) -
do_code_enrichment(bool) -
do_formula_enrichment(bool) -
force_backend_text(bool) -
table_structure_options(BaseTableStructureOptions) -
ocr_options(OcrOptions) -
layout_options(BaseLayoutOptions) -
images_scale(float) -
generate_page_images(bool) -
generate_picture_images(bool) -
generate_table_images(bool) -
generate_parsed_pages(bool) -
ocr_batch_size(int) -
layout_batch_size(int) -
table_batch_size(int) -
batch_polling_interval_seconds(float) -
queue_max_size(int)
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
batch_polling_interval_seconds: float = 0.5
pydantic-field
do_code_enrichment: bool = False
pydantic-field
do_formula_enrichment: bool = False
pydantic-field
do_ocr: bool = True
pydantic-field
do_picture_classification: bool = False
pydantic-field
do_picture_description: bool = False
pydantic-field
do_table_structure: bool = True
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
pydantic-field
force_backend_text: bool = False
pydantic-field
generate_page_images: bool = False
pydantic-field
generate_parsed_pages: bool = False
pydantic-field
generate_picture_images: bool = False
pydantic-field
generate_table_images: bool = False
pydantic-field
images_scale: float = 1.0
class-attribute
kind: str
pydantic-field
layout_batch_size: int = 4
pydantic-field
ocr_batch_size: int = 4
pydantic-field
picture_description_options: PictureDescriptionBaseOptions = smolvlm_picture_description
pydantic-field
queue_max_size: int = 100
pydantic-field
table_batch_size: int = 4
pydantic-model
Bases: PictureDescriptionBaseOptions
Show JSON schema:
{
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
},
"url": {
"default": "http://localhost:8000/v1/chat/completions",
"format": "uri",
"minLength": 1,
"title": "Url",
"type": "string"
},
"headers": {
"additionalProperties": {
"type": "string"
},
"default": {},
"title": "Headers",
"type": "object"
},
"params": {
"additionalProperties": true,
"default": {},
"title": "Params",
"type": "object"
},
"timeout": {
"default": 20,
"title": "Timeout",
"type": "number"
},
"concurrency": {
"default": 1,
"title": "Concurrency",
"type": "integer"
},
"prompt": {
"default": "Describe this image in a few sentences.",
"title": "Prompt",
"type": "string"
},
"provenance": {
"default": "",
"title": "Provenance",
"type": "string"
}
},
"title": "PictureDescriptionApiOptions",
"type": "object"
}
Fields:
-
batch_size(int) -
scale(float) -
picture_area_threshold(float) -
url(AnyUrl) -
headers(Dict[str, str]) -
params(Dict[str, Any]) -
timeout(float) -
concurrency(int) -
prompt(str) -
provenance(str)
pydantic-field
batch_size: int = 8
pydantic-field
concurrency: int = 1
pydantic-field
headers: Dict[str, str] = {}
class-attribute
kind: Literal['api'] = 'api'
pydantic-field
params: Dict[str, Any] = {}
pydantic-field
picture_area_threshold: float = 0.05
pydantic-field
prompt: str = 'Describe this image in a few sentences.'
pydantic-field
provenance: str = ''
pydantic-field
scale: float = 2
pydantic-field
timeout: float = 20
pydantic-field
url: AnyUrl
pydantic-model
Bases: BaseOptions
Show JSON schema:
{
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
}
Fields:
-
batch_size(int) -
scale(float) -
picture_area_threshold(float)
pydantic-field
batch_size: int = 8
class-attribute
kind: str
pydantic-field
picture_area_threshold: float = 0.05
pydantic-field
scale: float = 2
pydantic-model
Bases: PictureDescriptionBaseOptions
Show JSON schema:
{
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
},
"repo_id": {
"title": "Repo Id",
"type": "string"
},
"prompt": {
"default": "Describe this image in a few sentences.",
"title": "Prompt",
"type": "string"
},
"generation_config": {
"additionalProperties": true,
"default": {
"max_new_tokens": 200,
"do_sample": false
},
"title": "Generation Config",
"type": "object"
}
},
"required": [
"repo_id"
],
"title": "PictureDescriptionVlmOptions",
"type": "object"
}
Fields:
-
batch_size(int) -
scale(float) -
picture_area_threshold(float) -
repo_id(str) -
prompt(str) -
generation_config(Dict[str, Any])
pydantic-field
batch_size: int = 8
pydantic-field
generation_config: Dict[str, Any]
class-attribute
kind: Literal['vlm'] = 'vlm'
pydantic-field
picture_area_threshold: float = 0.05
pydantic-field
prompt: str = 'Describe this image in a few sentences.'
property
repo_cache_folder: str
pydantic-field
repo_id: str
pydantic-field
scale: float = 2
pydantic-model
Bases: BaseOptions
Base pipeline options.
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
}
},
"description": "Base pipeline options.",
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
}
},
"title": "PipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])])
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
class-attribute
kind: str
pydantic-model
Bases: OcrOptions
Options for the RapidOCR engine.
Show JSON schema:
{
"additionalProperties": false,
"description": "Options for the RapidOCR engine.",
"properties": {
"lang": {
"default": [
"english",
"chinese"
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
},
"backend": {
"default": "onnxruntime",
"enum": [
"onnxruntime",
"openvino",
"paddle",
"torch"
],
"title": "Backend",
"type": "string"
},
"text_score": {
"default": 0.5,
"title": "Text Score",
"type": "number"
},
"use_det": {
"anyOf": [
{
"type": "boolean"
},
{
"type": "null"
}
],
"default": null,
"title": "Use Det"
},
"use_cls": {
"anyOf": [
{
"type": "boolean"
},
{
"type": "null"
}
],
"default": null,
"title": "Use Cls"
},
"use_rec": {
"anyOf": [
{
"type": "boolean"
},
{
"type": "null"
}
],
"default": null,
"title": "Use Rec"
},
"print_verbose": {
"default": false,
"title": "Print Verbose",
"type": "boolean"
},
"det_model_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Det Model Path"
},
"cls_model_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Cls Model Path"
},
"rec_model_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Rec Model Path"
},
"rec_keys_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Rec Keys Path"
},
"rec_font_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Rec Font Path"
},
"font_path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Font Path"
},
"rapidocr_params": {
"additionalProperties": true,
"title": "Rapidocr Params",
"type": "object"
}
},
"title": "RapidOcrOptions",
"type": "object"
}
Config:
extra:forbid
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(List[str]) -
backend(Literal['onnxruntime', 'openvino', 'paddle', 'torch']) -
text_score(float) -
use_det(Optional[bool]) -
use_cls(Optional[bool]) -
use_rec(Optional[bool]) -
print_verbose(bool) -
det_model_path(Optional[str]) -
cls_model_path(Optional[str]) -
rec_model_path(Optional[str]) -
rec_keys_path(Optional[str]) -
rec_font_path(Optional[str]) -
font_path(Optional[str]) -
rapidocr_params(Dict[str, Any])
pydantic-field
backend: Literal['onnxruntime', 'openvino', 'paddle', 'torch'] = 'onnxruntime'
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
cls_model_path: Optional[str] = None
pydantic-field
det_model_path: Optional[str] = None
pydantic-field
font_path: Optional[str] = None
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: Literal['rapidocr'] = 'rapidocr'
pydantic-field
lang: List[str] = ['english', 'chinese']
class-attribute
instance-attribute
model_config = ConfigDict(extra='forbid')
pydantic-field
print_verbose: bool = False
pydantic-field
rapidocr_params: Dict[str, Any]
pydantic-field
rec_font_path: Optional[str] = None
pydantic-field
rec_keys_path: Optional[str] = None
pydantic-field
rec_model_path: Optional[str] = None
pydantic-field
text_score: float = 0.5
pydantic-field
use_cls: Optional[bool] = None
pydantic-field
use_det: Optional[bool] = None
pydantic-field
use_rec: Optional[bool] = None
pydantic-model
Bases: BaseTableStructureOptions
Options for the table structure.
Show JSON schema:
{
"$defs": {
"TableFormerMode": {
"description": "Modes for the TableFormer model.",
"enum": [
"fast",
"accurate"
],
"title": "TableFormerMode",
"type": "string"
}
},
"description": "Options for the table structure.",
"properties": {
"do_cell_matching": {
"default": true,
"title": "Do Cell Matching",
"type": "boolean"
},
"mode": {
"$ref": "#/$defs/TableFormerMode",
"default": "accurate"
}
},
"title": "TableStructureOptions",
"type": "object"
}
Fields:
-
do_cell_matching(bool) -
mode(TableFormerMode)
pydantic-field
do_cell_matching: bool = True
class-attribute
kind: str = 'docling_tableformer'
pydantic-model
Bases: OcrOptions
Options for the TesseractCli engine.
Show JSON schema:
{
"additionalProperties": false,
"description": "Options for the TesseractCli engine.",
"properties": {
"lang": {
"default": [
"fra",
"deu",
"spa",
"eng"
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
},
"tesseract_cmd": {
"default": "tesseract",
"title": "Tesseract Cmd",
"type": "string"
},
"path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Path"
},
"psm": {
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
],
"default": null,
"title": "Psm"
}
},
"title": "TesseractCliOcrOptions",
"type": "object"
}
Config:
extra:forbid
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(List[str]) -
tesseract_cmd(str) -
path(Optional[str]) -
psm(Optional[int])
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: Literal['tesseract'] = 'tesseract'
pydantic-field
lang: List[str] = ['fra', 'deu', 'spa', 'eng']
class-attribute
instance-attribute
model_config = ConfigDict(extra='forbid')
pydantic-field
path: Optional[str] = None
pydantic-field
psm: Optional[int] = None
pydantic-field
tesseract_cmd: str = 'tesseract'
pydantic-model
Bases: OcrOptions
Options for the Tesseract engine.
Show JSON schema:
{
"additionalProperties": false,
"description": "Options for the Tesseract engine.",
"properties": {
"lang": {
"default": [
"fra",
"deu",
"spa",
"eng"
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
},
"path": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Path"
},
"psm": {
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
],
"default": null,
"title": "Psm"
}
},
"title": "TesseractOcrOptions",
"type": "object"
}
Config:
extra:forbid
Fields:
-
force_full_page_ocr(Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])]) -
bitmap_area_threshold(Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])]) -
lang(List[str]) -
path(Optional[str]) -
psm(Optional[int])
pydantic-field
bitmap_area_threshold: Annotated[float, Field(description='Percentage of the page area for a bitmap to be processed with OCR.', examples=[0.05, 0.1])] = 0.05
pydantic-field
force_full_page_ocr: Annotated[bool, Field(description='If enabled, a full-page OCR is always applied.', examples=[False])] = False
class-attribute
kind: Literal['tesserocr'] = 'tesserocr'
pydantic-field
lang: List[str] = ['fra', 'deu', 'spa', 'eng']
class-attribute
instance-attribute
model_config = ConfigDict(extra='forbid')
pydantic-field
path: Optional[str] = None
pydantic-field
psm: Optional[int] = None
pydantic-model
Bases: PdfPipelineOptions
Pipeline options for the threaded PDF pipeline with batching and backpressure control
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"BaseLayoutOptions": {
"description": "Base options for layout models.",
"properties": {
"keep_empty_clusters": {
"default": false,
"title": "Keep Empty Clusters",
"type": "boolean"
},
"skip_cell_assignment": {
"default": false,
"title": "Skip Cell Assignment",
"type": "boolean"
}
},
"title": "BaseLayoutOptions",
"type": "object"
},
"BaseTableStructureOptions": {
"description": "Base options for table structure models.",
"properties": {},
"title": "BaseTableStructureOptions",
"type": "object"
},
"OcrOptions": {
"description": "OCR options.",
"properties": {
"lang": {
"description": "List of OCR languages to use. The format must match the values of the OCR engine of choice.",
"examples": [
[
"deu",
"eng"
]
],
"items": {
"type": "string"
},
"title": "Lang",
"type": "array"
},
"force_full_page_ocr": {
"default": false,
"description": "If enabled, a full-page OCR is always applied.",
"examples": [
false
],
"title": "Force Full Page Ocr",
"type": "boolean"
},
"bitmap_area_threshold": {
"default": 0.05,
"description": "Percentage of the page area for a bitmap to be processed with OCR.",
"examples": [
0.05,
0.1
],
"title": "Bitmap Area Threshold",
"type": "number"
}
},
"required": [
"lang"
],
"title": "OcrOptions",
"type": "object"
},
"PictureDescriptionBaseOptions": {
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
}
},
"description": "Pipeline options for the threaded PDF pipeline with batching and backpressure control",
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"do_picture_classification": {
"default": false,
"title": "Do Picture Classification",
"type": "boolean"
},
"do_picture_description": {
"default": false,
"title": "Do Picture Description",
"type": "boolean"
},
"picture_description_options": {
"$ref": "#/$defs/PictureDescriptionBaseOptions",
"default": {
"batch_size": 8,
"scale": 2.0,
"picture_area_threshold": 0.05,
"repo_id": "HuggingFaceTB/SmolVLM-256M-Instruct",
"prompt": "Describe this image in a few sentences.",
"generation_config": {
"do_sample": false,
"max_new_tokens": 200
}
}
},
"images_scale": {
"default": 1.0,
"title": "Images Scale",
"type": "number"
},
"generate_page_images": {
"default": false,
"title": "Generate Page Images",
"type": "boolean"
},
"generate_picture_images": {
"default": false,
"title": "Generate Picture Images",
"type": "boolean"
},
"do_table_structure": {
"default": true,
"title": "Do Table Structure",
"type": "boolean"
},
"do_ocr": {
"default": true,
"title": "Do Ocr",
"type": "boolean"
},
"do_code_enrichment": {
"default": false,
"title": "Do Code Enrichment",
"type": "boolean"
},
"do_formula_enrichment": {
"default": false,
"title": "Do Formula Enrichment",
"type": "boolean"
},
"force_backend_text": {
"default": false,
"title": "Force Backend Text",
"type": "boolean"
},
"table_structure_options": {
"$ref": "#/$defs/BaseTableStructureOptions",
"default": {
"do_cell_matching": true,
"mode": "accurate"
}
},
"ocr_options": {
"$ref": "#/$defs/OcrOptions",
"default": {
"lang": [],
"force_full_page_ocr": false,
"bitmap_area_threshold": 0.05
}
},
"layout_options": {
"$ref": "#/$defs/BaseLayoutOptions",
"default": {
"keep_empty_clusters": false,
"skip_cell_assignment": false,
"create_orphan_clusters": true,
"model_spec": {
"model_path": "",
"name": "docling_layout_heron",
"repo_id": "docling-project/docling-layout-heron",
"revision": "main",
"supported_devices": [
"cpu",
"cuda",
"mps"
]
}
}
},
"generate_table_images": {
"default": false,
"deprecated": true,
"title": "Generate Table Images",
"type": "boolean"
},
"generate_parsed_pages": {
"default": false,
"title": "Generate Parsed Pages",
"type": "boolean"
},
"ocr_batch_size": {
"default": 4,
"title": "Ocr Batch Size",
"type": "integer"
},
"layout_batch_size": {
"default": 4,
"title": "Layout Batch Size",
"type": "integer"
},
"table_batch_size": {
"default": 4,
"title": "Table Batch Size",
"type": "integer"
},
"batch_polling_interval_seconds": {
"default": 0.5,
"title": "Batch Polling Interval Seconds",
"type": "number"
},
"queue_max_size": {
"default": 100,
"title": "Queue Max Size",
"type": "integer"
}
},
"title": "ThreadedPdfPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
do_picture_classification(bool) -
do_picture_description(bool) -
picture_description_options(PictureDescriptionBaseOptions) -
images_scale(float) -
generate_page_images(bool) -
generate_picture_images(bool) -
do_table_structure(bool) -
do_ocr(bool) -
do_code_enrichment(bool) -
do_formula_enrichment(bool) -
force_backend_text(bool) -
table_structure_options(BaseTableStructureOptions) -
ocr_options(OcrOptions) -
layout_options(BaseLayoutOptions) -
generate_table_images(bool) -
generate_parsed_pages(bool) -
ocr_batch_size(int) -
layout_batch_size(int) -
table_batch_size(int) -
batch_polling_interval_seconds(float) -
queue_max_size(int)
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
batch_polling_interval_seconds: float = 0.5
pydantic-field
do_code_enrichment: bool = False
pydantic-field
do_formula_enrichment: bool = False
pydantic-field
do_ocr: bool = True
pydantic-field
do_picture_classification: bool = False
pydantic-field
do_picture_description: bool = False
pydantic-field
do_table_structure: bool = True
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
pydantic-field
force_backend_text: bool = False
pydantic-field
generate_page_images: bool = False
pydantic-field
generate_parsed_pages: bool = False
pydantic-field
generate_picture_images: bool = False
pydantic-field
generate_table_images: bool = False
pydantic-field
images_scale: float = 1.0
class-attribute
kind: str
pydantic-field
layout_batch_size: int = 4
pydantic-field
ocr_batch_size: int = 4
pydantic-field
picture_description_options: PictureDescriptionBaseOptions = smolvlm_picture_description
pydantic-field
queue_max_size: int = 100
pydantic-field
table_batch_size: int = 4
pydantic-model
Bases: PipelineOptions
Options for extraction pipeline.
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"InferenceFramework": {
"enum": [
"mlx",
"transformers",
"vllm"
],
"title": "InferenceFramework",
"type": "string"
},
"InlineVlmOptions": {
"properties": {
"kind": {
"const": "inline_model_options",
"default": "inline_model_options",
"title": "Kind",
"type": "string"
},
"prompt": {
"title": "Prompt",
"type": "string"
},
"scale": {
"default": 2.0,
"title": "Scale",
"type": "number"
},
"max_size": {
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
],
"default": null,
"title": "Max Size"
},
"temperature": {
"default": 0.0,
"title": "Temperature",
"type": "number"
},
"repo_id": {
"title": "Repo Id",
"type": "string"
},
"revision": {
"default": "main",
"title": "Revision",
"type": "string"
},
"trust_remote_code": {
"default": false,
"title": "Trust Remote Code",
"type": "boolean"
},
"load_in_8bit": {
"default": true,
"title": "Load In 8Bit",
"type": "boolean"
},
"llm_int8_threshold": {
"default": 6.0,
"title": "Llm Int8 Threshold",
"type": "number"
},
"quantized": {
"default": false,
"title": "Quantized",
"type": "boolean"
},
"inference_framework": {
"$ref": "#/$defs/InferenceFramework"
},
"transformers_model_type": {
"$ref": "#/$defs/TransformersModelType",
"default": "automodel"
},
"transformers_prompt_style": {
"$ref": "#/$defs/TransformersPromptStyle",
"default": "chat"
},
"response_format": {
"$ref": "#/$defs/ResponseFormat"
},
"torch_dtype": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Torch Dtype"
},
"supported_devices": {
"default": [
"cpu",
"cuda",
"mps"
],
"items": {
"$ref": "#/$defs/AcceleratorDevice"
},
"title": "Supported Devices",
"type": "array"
},
"stop_strings": {
"default": [],
"items": {
"type": "string"
},
"title": "Stop Strings",
"type": "array"
},
"custom_stopping_criteria": {
"default": [],
"items": {
"anyOf": []
},
"title": "Custom Stopping Criteria",
"type": "array"
},
"extra_generation_config": {
"additionalProperties": true,
"default": {},
"title": "Extra Generation Config",
"type": "object"
},
"extra_processor_kwargs": {
"additionalProperties": true,
"default": {},
"title": "Extra Processor Kwargs",
"type": "object"
},
"use_kv_cache": {
"default": true,
"title": "Use Kv Cache",
"type": "boolean"
},
"max_new_tokens": {
"default": 4096,
"title": "Max New Tokens",
"type": "integer"
},
"track_generated_tokens": {
"default": false,
"title": "Track Generated Tokens",
"type": "boolean"
},
"track_input_prompt": {
"default": false,
"title": "Track Input Prompt",
"type": "boolean"
}
},
"required": [
"prompt",
"repo_id",
"inference_framework",
"response_format"
],
"title": "InlineVlmOptions",
"type": "object"
},
"ResponseFormat": {
"enum": [
"doctags",
"markdown",
"html",
"otsl",
"plaintext"
],
"title": "ResponseFormat",
"type": "string"
},
"TransformersModelType": {
"enum": [
"automodel",
"automodel-vision2seq",
"automodel-causallm",
"automodel-imagetexttotext"
],
"title": "TransformersModelType",
"type": "string"
},
"TransformersPromptStyle": {
"enum": [
"chat",
"raw",
"none"
],
"title": "TransformersPromptStyle",
"type": "string"
}
},
"description": "Options for extraction pipeline.",
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"vlm_options": {
"$ref": "#/$defs/InlineVlmOptions",
"default": {
"kind": "inline_model_options",
"prompt": "",
"scale": 2.0,
"max_size": null,
"temperature": 0.0,
"repo_id": "numind/NuExtract-2.0-2B",
"revision": "fe5b2f0b63b81150721435a3ca1129a75c59c74e",
"trust_remote_code": false,
"load_in_8bit": true,
"llm_int8_threshold": 6.0,
"quantized": false,
"inference_framework": "transformers",
"transformers_model_type": "automodel-imagetexttotext",
"transformers_prompt_style": "chat",
"response_format": "plaintext",
"torch_dtype": "bfloat16",
"supported_devices": [
"cpu",
"cuda",
"mps"
],
"stop_strings": [],
"custom_stopping_criteria": [],
"extra_generation_config": {},
"extra_processor_kwargs": {},
"use_kv_cache": true,
"max_new_tokens": 4096,
"track_generated_tokens": false,
"track_input_prompt": false
}
}
},
"title": "VlmExtractionPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
vlm_options(Union[InlineVlmOptions])
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
class-attribute
kind: str
pydantic-field
vlm_options: Union[InlineVlmOptions] = NU_EXTRACT_2B_TRANSFORMERS
pydantic-model
Bases: PaginatedPipelineOptions
Show JSON schema:
{
"$defs": {
"AcceleratorDevice": {
"description": "Devices to run model inference",
"enum": [
"auto",
"cpu",
"cuda",
"mps"
],
"title": "AcceleratorDevice",
"type": "string"
},
"AcceleratorOptions": {
"additionalProperties": false,
"properties": {
"num_threads": {
"default": 4,
"title": "Num Threads",
"type": "integer"
},
"device": {
"anyOf": [
{
"type": "string"
},
{
"$ref": "#/$defs/AcceleratorDevice"
}
],
"default": "auto",
"title": "Device"
},
"cuda_use_flash_attention2": {
"default": false,
"title": "Cuda Use Flash Attention2",
"type": "boolean"
}
},
"title": "AcceleratorOptions",
"type": "object"
},
"InferenceFramework": {
"enum": [
"mlx",
"transformers",
"vllm"
],
"title": "InferenceFramework",
"type": "string"
},
"InlineVlmOptions": {
"properties": {
"kind": {
"const": "inline_model_options",
"default": "inline_model_options",
"title": "Kind",
"type": "string"
},
"prompt": {
"title": "Prompt",
"type": "string"
},
"scale": {
"default": 2.0,
"title": "Scale",
"type": "number"
},
"max_size": {
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
],
"default": null,
"title": "Max Size"
},
"temperature": {
"default": 0.0,
"title": "Temperature",
"type": "number"
},
"repo_id": {
"title": "Repo Id",
"type": "string"
},
"revision": {
"default": "main",
"title": "Revision",
"type": "string"
},
"trust_remote_code": {
"default": false,
"title": "Trust Remote Code",
"type": "boolean"
},
"load_in_8bit": {
"default": true,
"title": "Load In 8Bit",
"type": "boolean"
},
"llm_int8_threshold": {
"default": 6.0,
"title": "Llm Int8 Threshold",
"type": "number"
},
"quantized": {
"default": false,
"title": "Quantized",
"type": "boolean"
},
"inference_framework": {
"$ref": "#/$defs/InferenceFramework"
},
"transformers_model_type": {
"$ref": "#/$defs/TransformersModelType",
"default": "automodel"
},
"transformers_prompt_style": {
"$ref": "#/$defs/TransformersPromptStyle",
"default": "chat"
},
"response_format": {
"$ref": "#/$defs/ResponseFormat"
},
"torch_dtype": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Torch Dtype"
},
"supported_devices": {
"default": [
"cpu",
"cuda",
"mps"
],
"items": {
"$ref": "#/$defs/AcceleratorDevice"
},
"title": "Supported Devices",
"type": "array"
},
"stop_strings": {
"default": [],
"items": {
"type": "string"
},
"title": "Stop Strings",
"type": "array"
},
"custom_stopping_criteria": {
"default": [],
"items": {
"anyOf": []
},
"title": "Custom Stopping Criteria",
"type": "array"
},
"extra_generation_config": {
"additionalProperties": true,
"default": {},
"title": "Extra Generation Config",
"type": "object"
},
"extra_processor_kwargs": {
"additionalProperties": true,
"default": {},
"title": "Extra Processor Kwargs",
"type": "object"
},
"use_kv_cache": {
"default": true,
"title": "Use Kv Cache",
"type": "boolean"
},
"max_new_tokens": {
"default": 4096,
"title": "Max New Tokens",
"type": "integer"
},
"track_generated_tokens": {
"default": false,
"title": "Track Generated Tokens",
"type": "boolean"
},
"track_input_prompt": {
"default": false,
"title": "Track Input Prompt",
"type": "boolean"
}
},
"required": [
"prompt",
"repo_id",
"inference_framework",
"response_format"
],
"title": "InlineVlmOptions",
"type": "object"
},
"PictureDescriptionBaseOptions": {
"properties": {
"batch_size": {
"default": 8,
"title": "Batch Size",
"type": "integer"
},
"scale": {
"default": 2,
"title": "Scale",
"type": "number"
},
"picture_area_threshold": {
"default": 0.05,
"title": "Picture Area Threshold",
"type": "number"
}
},
"title": "PictureDescriptionBaseOptions",
"type": "object"
},
"ResponseFormat": {
"enum": [
"doctags",
"markdown",
"html",
"otsl",
"plaintext"
],
"title": "ResponseFormat",
"type": "string"
},
"TransformersModelType": {
"enum": [
"automodel",
"automodel-vision2seq",
"automodel-causallm",
"automodel-imagetexttotext"
],
"title": "TransformersModelType",
"type": "string"
},
"TransformersPromptStyle": {
"enum": [
"chat",
"raw",
"none"
],
"title": "TransformersPromptStyle",
"type": "string"
}
},
"properties": {
"document_timeout": {
"anyOf": [
{
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"description": "Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.",
"examples": [
10.0,
20.0
],
"title": "Document Timeout"
},
"accelerator_options": {
"$ref": "#/$defs/AcceleratorOptions",
"default": {
"num_threads": 4,
"device": "auto",
"cuda_use_flash_attention2": false
},
"description": "Configuration options for hardware acceleration (e.g., GPU or optimized execution settings)."
},
"enable_remote_services": {
"default": false,
"description": "Enable calling external APIs or cloud services during pipeline execution.",
"examples": [
false
],
"title": "Enable Remote Services",
"type": "boolean"
},
"allow_external_plugins": {
"default": false,
"description": "Allow loading external third-party plugins or modules. Disabled by default for safety.",
"examples": [
false
],
"title": "Allow External Plugins",
"type": "boolean"
},
"artifacts_path": {
"anyOf": [
{
"format": "path",
"type": "string"
},
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.",
"examples": [
"./artifacts",
"/tmp/docling_outputs"
],
"title": "Artifacts Path"
},
"do_picture_classification": {
"default": false,
"title": "Do Picture Classification",
"type": "boolean"
},
"do_picture_description": {
"default": false,
"title": "Do Picture Description",
"type": "boolean"
},
"picture_description_options": {
"$ref": "#/$defs/PictureDescriptionBaseOptions",
"default": {
"batch_size": 8,
"scale": 2.0,
"picture_area_threshold": 0.05,
"repo_id": "HuggingFaceTB/SmolVLM-256M-Instruct",
"prompt": "Describe this image in a few sentences.",
"generation_config": {
"do_sample": false,
"max_new_tokens": 200
}
}
},
"images_scale": {
"default": 1.0,
"title": "Images Scale",
"type": "number"
},
"generate_page_images": {
"default": true,
"title": "Generate Page Images",
"type": "boolean"
},
"generate_picture_images": {
"default": false,
"title": "Generate Picture Images",
"type": "boolean"
},
"force_backend_text": {
"default": false,
"title": "Force Backend Text",
"type": "boolean"
},
"vlm_options": {
"$ref": "#/$defs/InlineVlmOptions",
"default": {
"kind": "inline_model_options",
"prompt": "Convert this page to docling.",
"scale": 2.0,
"max_size": null,
"temperature": 0.0,
"repo_id": "ibm-granite/granite-docling-258M",
"revision": "main",
"trust_remote_code": false,
"load_in_8bit": true,
"llm_int8_threshold": 6.0,
"quantized": false,
"inference_framework": "transformers",
"transformers_model_type": "automodel-imagetexttotext",
"transformers_prompt_style": "chat",
"response_format": "doctags",
"torch_dtype": null,
"supported_devices": [
"cpu",
"cuda"
],
"stop_strings": [
"</doctag>",
"<|end_of_text|>"
],
"custom_stopping_criteria": [],
"extra_generation_config": {
"skip_special_tokens": false
},
"extra_processor_kwargs": {},
"use_kv_cache": true,
"max_new_tokens": 8192,
"track_generated_tokens": false,
"track_input_prompt": false
},
"title": "Vlm Options"
}
},
"title": "VlmPipelineOptions",
"type": "object"
}
Fields:
-
document_timeout(Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])]) -
accelerator_options(Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]) -
enable_remote_services(Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])]) -
allow_external_plugins(Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])]) -
artifacts_path(Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])]) -
do_picture_classification(bool) -
do_picture_description(bool) -
picture_description_options(PictureDescriptionBaseOptions) -
images_scale(float) -
generate_picture_images(bool) -
generate_page_images(bool) -
force_backend_text(bool) -
vlm_options(Union[InlineVlmOptions, ApiVlmOptions])
pydantic-field
accelerator_options: Annotated[AcceleratorOptions, Field(description='Configuration options for hardware acceleration (e.g., GPU or optimized execution settings).')]
pydantic-field
allow_external_plugins: Annotated[bool, Field(description='Allow loading external third-party plugins or modules. Disabled by default for safety.', examples=[False])] = False
pydantic-field
artifacts_path: Annotated[Optional[Union[Path, str]], Field(description='Filesystem path where pipeline artifacts should be stored. If None, artifacts will be fetched. You can use the utility `docling-tools models download` to pre-fetch the model artifacts.', examples=['./artifacts', '/tmp/docling_outputs'])] = None
pydantic-field
do_picture_classification: bool = False
pydantic-field
do_picture_description: bool = False
pydantic-field
document_timeout: Annotated[Optional[float], Field(description='Maximum allowed processing time for a document before timing out. If None, no timeout is enforced.', examples=[10.0, 20.0])] = None
pydantic-field
enable_remote_services: Annotated[bool, Field(description='Enable calling external APIs or cloud services during pipeline execution.', examples=[False])] = False
pydantic-field
force_backend_text: bool = False
pydantic-field
generate_page_images: bool = True
pydantic-field
generate_picture_images: bool = False
pydantic-field
images_scale: float = 1.0
class-attribute
kind: str
pydantic-field
picture_description_options: PictureDescriptionBaseOptions = smolvlm_picture_description
pydantic-field
vlm_options: Union[InlineVlmOptions, ApiVlmOptions] = GRANITEDOCLING_TRANSFORMERS