docling 2.40.0__tar.gz → 2.41.0__tar.gz

{docling-2.40.0 → docling-2.41.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling
-Version: 2.40.0
+Version: 2.41.0
 Summary: SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications.
 Author-email: Christoph Auer <cau@zurich.ibm.com>, Michele Dolfi <dol@zurich.ibm.com>, Maxim Lysak <mly@zurich.ibm.com>, Nikos Livathinos <nli@zurich.ibm.com>, Ahmed Nassar <ahn@zurich.ibm.com>, Panos Vagenas <pva@zurich.ibm.com>, Peter Staar <taa@zurich.ibm.com>
 License-Expression: MIT
@@ -26,7 +26,7 @@ Requires-Python: <4.0,>=3.9
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: pydantic<3.0.0,>=2.0.0
-Requires-Dist: docling-core[chunking]<3.0.0,>=2.39.0
+Requires-Dist: docling-core[chunking]<3.0.0,>=2.42.0
 Requires-Dist: docling-parse<5.0.0,>=4.0.0
 Requires-Dist: docling-ibm-models<4,>=3.6.0
 Requires-Dist: filetype<2.0.0,>=1.2.0
@@ -62,7 +62,7 @@ Provides-Extra: rapidocr
 Requires-Dist: rapidocr-onnxruntime<2.0.0,>=1.4.0; python_version < "3.13" and extra == "rapidocr"
 Requires-Dist: onnxruntime<2.0.0,>=1.7.0; extra == "rapidocr"
 Provides-Extra: asr
-Requires-Dist: openai-whisper>=20240930; extra == "asr"
+Requires-Dist: openai-whisper>=20250625; extra == "asr"
 Dynamic: license-file
 
 <p align="center">

{docling-2.40.0 → docling-2.41.0}/docling/datamodel/asr_model_specs.py

@@ -22,7 +22,7 @@ WHISPER_TINY = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )
@@ -33,7 +33,7 @@ WHISPER_SMALL = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )
@@ -44,7 +44,7 @@ WHISPER_MEDIUM = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )
@@ -55,7 +55,7 @@ WHISPER_BASE = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )
@@ -66,7 +66,7 @@ WHISPER_LARGE = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )
@@ -77,7 +77,7 @@ WHISPER_TURBO = InlineAsrNativeWhisperOptions(
     verbose=True,
     timestamps=True,
     word_timestamps=True,
-    temperatue=0.0,
+    temperature=0.0,
     max_new_tokens=256,
     max_time_chunk=30.0,
 )

{docling-2.40.0 → docling-2.41.0}/docling/datamodel/base_models.py

@@ -12,6 +12,7 @@ from docling_core.types.doc import (
     Size,
     TableCell,
 )
+from docling_core.types.doc.base import PydanticSerCtxKey, round_pydantic_float
 from docling_core.types.doc.page import SegmentedPdfPage, TextCell
 from docling_core.types.io import (
     DocumentStream,
@@ -19,7 +20,14 @@ from docling_core.types.io import (
 
 # DO NOT REMOVE; explicitly exposed from this location
 from PIL.Image import Image
-from pydantic import BaseModel, ConfigDict, Field, computed_field
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    FieldSerializationInfo,
+    computed_field,
+    field_serializer,
+)
 
 if TYPE_CHECKING:
     from docling.backend.pdf_backend import PdfPageBackend
@@ -142,6 +150,10 @@ class Cluster(BaseModel):
     cells: List[TextCell] = []
     children: List["Cluster"] = []  # Add child cluster support
 
+    @field_serializer("confidence")
+    def _serialize(self, value: float, info: FieldSerializationInfo) -> float:
+        return round_pydantic_float(value, info.context, PydanticSerCtxKey.CONFID_PREC)
+
 
 class BasePageElement(BaseModel):
     label: DocItemLabel
@@ -194,6 +206,16 @@ class FigureElement(BasePageElement):
     predicted_class: Optional[str] = None
     confidence: Optional[float] = None
 
+    @field_serializer("confidence")
+    def _serialize(
+        self, value: Optional[float], info: FieldSerializationInfo
+    ) -> Optional[float]:
+        return (
+            round_pydantic_float(value, info.context, PydanticSerCtxKey.CONFID_PREC)
+            if value is not None
+            else None
+        )
+
 
 class FigureClassificationPrediction(BaseModel):
     figure_count: int = 0

docling-2.41.0/docling/datamodel/layout_model_specs.py

@@ -0,0 +1,90 @@
+import logging
+from enum import Enum
+from pathlib import Path
+from typing import Optional
+
+from pydantic import BaseModel
+
+from docling.datamodel.accelerator_options import AcceleratorDevice
+
+_log = logging.getLogger(__name__)
+
+
+class LayoutModelConfig(BaseModel):
+    name: str
+    repo_id: str
+    revision: str
+    model_path: str
+    supported_devices: list[AcceleratorDevice] = [
+        AcceleratorDevice.CPU,
+        AcceleratorDevice.CUDA,
+        AcceleratorDevice.MPS,
+    ]
+
+    @property
+    def model_repo_folder(self) -> str:
+        return self.repo_id.replace("/", "--")
+
+
+# HuggingFace Layout Models
+
+# Default Docling Layout Model
+DOCLING_LAYOUT_V2 = LayoutModelConfig(
+    name="docling_layout_v2",
+    repo_id="ds4sd/docling-layout-old",
+    revision="main",
+    model_path="",
+)
+
+DOCLING_LAYOUT_HERON = LayoutModelConfig(
+    name="docling_layout_heron",
+    repo_id="ds4sd/docling-layout-heron",
+    revision="main",
+    model_path="",
+)
+
+DOCLING_LAYOUT_HERON_101 = LayoutModelConfig(
+    name="docling_layout_heron_101",
+    repo_id="ds4sd/docling-layout-heron-101",
+    revision="main",
+    model_path="",
+)
+
+DOCLING_LAYOUT_EGRET_MEDIUM = LayoutModelConfig(
+    name="docling_layout_egret_medium",
+    repo_id="ds4sd/docling-layout-egret-medium",
+    revision="main",
+    model_path="",
+)
+
+DOCLING_LAYOUT_EGRET_LARGE = LayoutModelConfig(
+    name="docling_layout_egret_large",
+    repo_id="ds4sd/docling-layout-egret-large",
+    revision="main",
+    model_path="",
+)
+
+DOCLING_LAYOUT_EGRET_XLARGE = LayoutModelConfig(
+    name="docling_layout_egret_xlarge",
+    repo_id="ds4sd/docling-layout-egret-xlarge",
+    revision="main",
+    model_path="",
+)
+
+# Example for a hypothetical alternative model
+# ALTERNATIVE_LAYOUT = LayoutModelConfig(
+#     name="alternative_layout",
+#     repo_id="someorg/alternative-layout",
+#     revision="main",
+#     model_path="model_artifacts/layout_alt",
+# )
+
+
+class LayoutModelType(str, Enum):
+    DOCLING_LAYOUT_V2 = "docling_layout_v2"
+    DOCLING_LAYOUT_HERON = "docling_layout_heron"
+    DOCLING_LAYOUT_HERON_101 = "docling_layout_heron_101"
+    DOCLING_LAYOUT_EGRET_MEDIUM = "docling_layout_egret_medium"
+    DOCLING_LAYOUT_EGRET_LARGE = "docling_layout_egret_large"
+    DOCLING_LAYOUT_EGRET_XLARGE = "docling_layout_egret_xlarge"
+    # ALTERNATIVE_LAYOUT = "alternative_layout"

{docling-2.40.0 → docling-2.41.0}/docling/datamodel/pipeline_options.py

@@ -16,6 +16,15 @@ from docling.datamodel import asr_model_specs
 
 # Import the following for backwards compatibility
 from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
+from docling.datamodel.layout_model_specs import (
+    DOCLING_LAYOUT_EGRET_LARGE,
+    DOCLING_LAYOUT_EGRET_MEDIUM,
+    DOCLING_LAYOUT_EGRET_XLARGE,
+    DOCLING_LAYOUT_HERON,
+    DOCLING_LAYOUT_HERON_101,
+    DOCLING_LAYOUT_V2,
+    LayoutModelConfig,
+)
 from docling.datamodel.pipeline_options_asr_model import (
     InlineAsrOptions,
 )
@@ -270,6 +279,7 @@ class LayoutOptions(BaseModel):
     """Options for layout processing."""
 
     create_orphan_clusters: bool = True  # Whether to create clusters for orphaned cells
+    model_spec: LayoutModelConfig = DOCLING_LAYOUT_V2
 
 
 class AsrPipelineOptions(PipelineOptions):

{docling-2.40.0 → docling-2.41.0}/docling/datamodel/pipeline_options_vlm_model.py

@@ -1,6 +1,7 @@
 from enum import Enum
-from typing import Any, Dict, List, Literal, Optional, Union
+from typing import Any, Callable, Dict, List, Literal, Optional, Union
 
+from docling_core.types.doc.page import SegmentedPage
 from pydantic import AnyUrl, BaseModel
 from typing_extensions import deprecated
 
@@ -9,9 +10,10 @@ from docling.datamodel.accelerator_options import AcceleratorDevice
 
 class BaseVlmOptions(BaseModel):
     kind: str
-    prompt: str
+    prompt: Union[str, Callable[[Optional[SegmentedPage]], str]]
     scale: float = 2.0
     max_size: Optional[int] = None
+    temperature: float = 0.0
 
 
 class ResponseFormat(str, Enum):
@@ -29,6 +31,12 @@ class TransformersModelType(str, Enum):
     AUTOMODEL = "automodel"
     AUTOMODEL_VISION2SEQ = "automodel-vision2seq"
     AUTOMODEL_CAUSALLM = "automodel-causallm"
+    AUTOMODEL_IMAGETEXTTOTEXT = "automodel-imagetexttotext"
+
+
+class TransformersPromptStyle(str, Enum):
+    CHAT = "chat"
+    RAW = "raw"
 
 
 class InlineVlmOptions(BaseVlmOptions):
@@ -42,6 +50,7 @@ class InlineVlmOptions(BaseVlmOptions):
 
     inference_framework: InferenceFramework
     transformers_model_type: TransformersModelType = TransformersModelType.AUTOMODEL
+    transformers_prompt_style: TransformersPromptStyle = TransformersPromptStyle.CHAT
     response_format: ResponseFormat
 
     torch_dtype: Optional[str] = None
@@ -51,7 +60,6 @@ class InlineVlmOptions(BaseVlmOptions):
         AcceleratorDevice.MPS,
     ]
 
-    temperature: float = 0.0
     stop_strings: List[str] = []
     extra_generation_config: Dict[str, Any] = {}
 

{docling-2.40.0 → docling-2.41.0}/docling/models/api_vlm_model.py

@@ -29,12 +29,9 @@ class ApiVlmModel(BasePageModel):
 
             self.timeout = self.vlm_options.timeout
             self.concurrency = self.vlm_options.concurrency
-            self.prompt_content = (
-                f"This is a page from a document.\n{self.vlm_options.prompt}"
-            )
             self.params = {
                 **self.vlm_options.params,
-                "temperature": 0,
+                "temperature": self.vlm_options.temperature,
             }
 
     def __call__(
@@ -56,9 +53,14 @@ class ApiVlmModel(BasePageModel):
                         if hi_res_image.mode != "RGB":
                             hi_res_image = hi_res_image.convert("RGB")
 
+                    if callable(self.vlm_options.prompt):
+                        prompt = self.vlm_options.prompt(page.parsed_page)
+                    else:
+                        prompt = self.vlm_options.prompt
+
                     page_tags = api_image_request(
                         image=hi_res_image,
-                        prompt=self.prompt_content,
+                        prompt=prompt,
                         url=self.vlm_options.url,
                         timeout=self.timeout,
                         headers=self.vlm_options.headers,

{docling-2.40.0 → docling-2.41.0}/docling/models/document_picture_classifier.py

@@ -14,7 +14,8 @@ from PIL import Image
 from pydantic import BaseModel
 
 from docling.datamodel.accelerator_options import AcceleratorOptions
-from docling.models.base_model import BaseEnrichmentModel
+from docling.datamodel.base_models import ItemAndImageEnrichmentElement
+from docling.models.base_model import BaseItemAndImageEnrichmentModel
 from docling.models.utils.hf_model_download import download_hf_model
 from docling.utils.accelerator_utils import decide_device
 
@@ -32,7 +33,7 @@ class DocumentPictureClassifierOptions(BaseModel):
     kind: Literal["document_picture_classifier"] = "document_picture_classifier"
 
 
-class DocumentPictureClassifier(BaseEnrichmentModel):
+class DocumentPictureClassifier(BaseItemAndImageEnrichmentModel):
     """
     A model for classifying pictures in documents.
 
@@ -135,7 +136,7 @@ class DocumentPictureClassifier(BaseEnrichmentModel):
     def __call__(
         self,
         doc: DoclingDocument,
-        element_batch: Iterable[NodeItem],
+        element_batch: Iterable[ItemAndImageEnrichmentElement],
     ) -> Iterable[NodeItem]:
         """
         Processes a batch of elements and enriches them with classification predictions.
@@ -144,7 +145,7 @@ class DocumentPictureClassifier(BaseEnrichmentModel):
         ----------
         doc : DoclingDocument
             The document containing the elements to be processed.
-        element_batch : Iterable[NodeItem]
+        element_batch : Iterable[ItemAndImageEnrichmentElement]
             A batch of pictures to classify.
 
         Returns
@@ -155,22 +156,20 @@ class DocumentPictureClassifier(BaseEnrichmentModel):
         """
         if not self.enabled:
             for element in element_batch:
-                yield element
+                yield element.item
             return
 
         images: List[Union[Image.Image, np.ndarray]] = []
         elements: List[PictureItem] = []
         for el in element_batch:
-            assert isinstance(el, PictureItem)
-            elements.append(el)
-            img = el.get_image(doc)
-            assert img is not None
-            images.append(img)
+            assert isinstance(el.item, PictureItem)
+            elements.append(el.item)
+            images.append(el.image)
 
         outputs = self.document_picture_classifier.predict(images)
 
-        for element, output in zip(elements, outputs):
-            element.annotations.append(
+        for item, output in zip(elements, outputs):
+            item.annotations.append(
                 PictureClassificationData(
                     provenance="DocumentPictureClassifier",
                     predicted_classes=[
@@ -183,4 +182,4 @@ class DocumentPictureClassifier(BaseEnrichmentModel):
                 )
             )
 
-            yield element
+            yield item

{docling-2.40.0 → docling-2.41.0}/docling/models/layout_model.py

@@ -12,6 +12,7 @@ from PIL import Image
 from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import BoundingBox, Cluster, LayoutPrediction, Page
 from docling.datamodel.document import ConversionResult
+from docling.datamodel.layout_model_specs import DOCLING_LAYOUT_V2, LayoutModelConfig
 from docling.datamodel.pipeline_options import LayoutOptions
 from docling.datamodel.settings import settings
 from docling.models.base_model import BasePageModel
@@ -25,9 +26,6 @@ _log = logging.getLogger(__name__)
 
 
 class LayoutModel(BasePageModel):
-    _model_repo_folder = "ds4sd--docling-models"
-    _model_path = "model_artifacts/layout"
-
     TEXT_ELEM_LABELS = [
         DocItemLabel.TEXT,
         DocItemLabel.FOOTNOTE,
@@ -59,25 +57,28 @@ class LayoutModel(BasePageModel):
         self.options = options
 
         device = decide_device(accelerator_options.device)
+        layout_model_config = options.model_spec
+        model_repo_folder = layout_model_config.model_repo_folder
+        model_path = layout_model_config.model_path
 
         if artifacts_path is None:
-            artifacts_path = self.download_models() / self._model_path
+            artifacts_path = (
+                self.download_models(layout_model_config=layout_model_config)
+                / model_path
+            )
         else:
-            # will become the default in the future
-            if (artifacts_path / self._model_repo_folder).exists():
-                artifacts_path = (
-                    artifacts_path / self._model_repo_folder / self._model_path
-                )
-            elif (artifacts_path / self._model_path).exists():
+            if (artifacts_path / model_repo_folder).exists():
+                artifacts_path = artifacts_path / model_repo_folder / model_path
+            elif (artifacts_path / model_path).exists():
                 warnings.warn(
                     "The usage of artifacts_path containing directly "
-                    f"{self._model_path} is deprecated. Please point "
+                    f"{model_path} is deprecated. Please point "
                     "the artifacts_path to the parent containing "
-                    f"the {self._model_repo_folder} folder.",
+                    f"the {model_repo_folder} folder.",
                     DeprecationWarning,
                     stacklevel=3,
                 )
-                artifacts_path = artifacts_path / self._model_path
+                artifacts_path = artifacts_path / model_path
 
         self.layout_predictor = LayoutPredictor(
             artifact_path=str(artifacts_path),
@@ -90,10 +91,11 @@ class LayoutModel(BasePageModel):
         local_dir: Optional[Path] = None,
         force: bool = False,
         progress: bool = False,
+        layout_model_config: LayoutModelConfig = DOCLING_LAYOUT_V2,
     ) -> Path:
         return download_hf_model(
-            repo_id="ds4sd/docling-models",
-            revision="v2.2.0",
+            repo_id=layout_model_config.repo_id,
+            revision=layout_model_config.revision,
             local_dir=local_dir,
             force=force,
             progress=progress,

{docling-2.40.0 → docling-2.41.0}/docling/models/vlm_models_inline/hf_transformers_model.py

@@ -13,6 +13,7 @@ from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options_vlm_model import (
     InlineVlmOptions,
     TransformersModelType,
+    TransformersPromptStyle,
 )
 from docling.models.base_model import BasePageModel
 from docling.models.utils.hf_model_download import (
@@ -41,6 +42,7 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
             from transformers import (
                 AutoModel,
                 AutoModelForCausalLM,
+                AutoModelForImageTextToText,
                 AutoModelForVision2Seq,
                 AutoProcessor,
                 BitsAndBytesConfig,
@@ -91,6 +93,11 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
                 == TransformersModelType.AUTOMODEL_VISION2SEQ
             ):
                 model_cls = AutoModelForVision2Seq
+            elif (
+                self.vlm_options.transformers_model_type
+                == TransformersModelType.AUTOMODEL_IMAGETEXTTOTEXT
+            ):
+                model_cls = AutoModelForImageTextToText
 
             self.processor = AutoProcessor.from_pretrained(
                 artifacts_path,
@@ -128,7 +135,11 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
                     )
 
                     # Define prompt structure
-                    prompt = self.formulate_prompt()
+                    if callable(self.vlm_options.prompt):
+                        user_prompt = self.vlm_options.prompt(page.parsed_page)
+                    else:
+                        user_prompt = self.vlm_options.prompt
+                    prompt = self.formulate_prompt(user_prompt)
 
                     inputs = self.processor(
                         text=prompt, images=[hi_res_image], return_tensors="pt"
@@ -162,10 +173,13 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
 
                 yield page
 
-    def formulate_prompt(self) -> str:
+    def formulate_prompt(self, user_prompt: str) -> str:
         """Formulate a prompt for the VLM."""
 
-        if self.vlm_options.repo_id == "microsoft/Phi-4-multimodal-instruct":
+        if self.vlm_options.transformers_prompt_style == TransformersPromptStyle.RAW:
+            return user_prompt
+
+        elif self.vlm_options.repo_id == "microsoft/Phi-4-multimodal-instruct":
             _log.debug("Using specialized prompt for Phi-4")
             # more info here: https://huggingface.co/microsoft/Phi-4-multimodal-instruct#loading-the-model-locally
 
@@ -173,25 +187,30 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
             assistant_prompt = "<|assistant|>"
             prompt_suffix = "<|end|>"
 
-            prompt = f"{user_prompt}<|image_1|>{self.vlm_options.prompt}{prompt_suffix}{assistant_prompt}"
+            prompt = f"{user_prompt}<|image_1|>{user_prompt}{prompt_suffix}{assistant_prompt}"
             _log.debug(f"prompt for {self.vlm_options.repo_id}: {prompt}")
 
             return prompt
 
-        messages = [
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "text",
-                        "text": "This is a page from a document.",
-                    },
-                    {"type": "image"},
-                    {"type": "text", "text": self.vlm_options.prompt},
-                ],
-            }
-        ]
-        prompt = self.processor.apply_chat_template(
-            messages, add_generation_prompt=False
+        elif self.vlm_options.transformers_prompt_style == TransformersPromptStyle.CHAT:
+            messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "text",
+                            "text": "This is a page from a document.",
+                        },
+                        {"type": "image"},
+                        {"type": "text", "text": user_prompt},
+                    ],
+                }
+            ]
+            prompt = self.processor.apply_chat_template(
+                messages, add_generation_prompt=False
+            )
+            return prompt
+
+        raise RuntimeError(
+            f"Uknown prompt style `{self.vlm_options.transformers_prompt_style}`. Valid values are {', '.join(s.value for s in TransformersPromptStyle)}."
         )
-        return prompt

{docling-2.40.0 → docling-2.41.0}/docling/models/vlm_models_inline/mlx_model.py

@@ -56,8 +56,6 @@ class HuggingFaceMlxModel(BasePageModel, HuggingFaceModelDownloadMixin):
             elif (artifacts_path / repo_cache_folder).exists():
                 artifacts_path = artifacts_path / repo_cache_folder
 
-            self.param_question = vlm_options.prompt
-
             ## Load the model
             self.vlm_model, self.processor = load(artifacts_path)
             self.config = load_config(artifacts_path)
@@ -86,8 +84,12 @@ class HuggingFaceMlxModel(BasePageModel, HuggingFaceModelDownloadMixin):
                         if hi_res_image.mode != "RGB":
                             hi_res_image = hi_res_image.convert("RGB")
 
+                    if callable(self.vlm_options.prompt):
+                        user_prompt = self.vlm_options.prompt(page.parsed_page)
+                    else:
+                        user_prompt = self.vlm_options.prompt
                     prompt = self.apply_chat_template(
-                        self.processor, self.config, self.param_question, num_images=1
+                        self.processor, self.config, user_prompt, num_images=1
                     )
 
                     start_time = time.time()

{docling-2.40.0 → docling-2.41.0}/docling/pipeline/standard_pdf_pipeline.py

@@ -10,6 +10,7 @@ from docling.backend.abstract_backend import AbstractDocumentBackend
 from docling.backend.pdf_backend import PdfDocumentBackend
 from docling.datamodel.base_models import AssembledUnit, Page
 from docling.datamodel.document import ConversionResult
+from docling.datamodel.layout_model_specs import LayoutModelConfig
 from docling.datamodel.pipeline_options import PdfPipelineOptions
 from docling.datamodel.settings import settings
 from docling.models.base_ocr_model import BaseOcrModel
@@ -36,9 +37,6 @@ _log = logging.getLogger(__name__)
 
 
 class StandardPdfPipeline(PaginatedPipeline):
-    _layout_model_path = LayoutModel._model_path
-    _table_model_path = TableStructureModel._model_path
-
     def __init__(self, pipeline_options: PdfPipelineOptions):
         super().__init__(pipeline_options)
         self.pipeline_options: PdfPipelineOptions
@@ -129,6 +127,7 @@ class StandardPdfPipeline(PaginatedPipeline):
         if (
             self.pipeline_options.do_formula_enrichment
             or self.pipeline_options.do_code_enrichment
+            or self.pipeline_options.do_picture_classification
             or self.pipeline_options.do_picture_description
         ):
             self.keep_backend = True

{docling-2.40.0 → docling-2.41.0}/docling/pipeline/vlm_pipeline.py

@@ -117,6 +117,7 @@ class VlmPipeline(PaginatedPipeline):
             page._backend = conv_res.input._backend.load_page(page.page_no)  # type: ignore
             if page._backend is not None and page._backend.is_valid():
                 page.size = page._backend.get_size()
+                page.parsed_page = page._backend.get_segmented_page()
 
         return page
 

{docling-2.40.0 → docling-2.41.0}/docling/utils/model_downloader.py

@@ -2,6 +2,7 @@ import logging
 from pathlib import Path
 from typing import Optional
 
+from docling.datamodel.layout_model_specs import DOCLING_LAYOUT_V2
 from docling.datamodel.pipeline_options import (
     granite_picture_description,
     smolvlm_picture_description,
@@ -46,7 +47,7 @@ def download_models(
     if with_layout:
         _log.info("Downloading layout model...")
         LayoutModel.download_models(
-            local_dir=output_dir / LayoutModel._model_repo_folder,
+            local_dir=output_dir / DOCLING_LAYOUT_V2.model_repo_folder,
             force=force,
             progress=progress,
         )

{docling-2.40.0 → docling-2.41.0}/docling/utils/ocr_utils.py

@@ -41,7 +41,7 @@ def tesseract_box_to_bounding_rectangle(
     im_size: Tuple[int, int],
 ) -> BoundingRectangle:
     # box is in the top, left, height, width format, top left coordinates
-    rect = rotate_bounding_box(bbox, angle=-orientation, im_size=im_size)
+    rect = rotate_bounding_box(bbox, angle=orientation, im_size=im_size)
     rect = BoundingRectangle(
         r_x0=rect.r_x0 / scale,
         r_y0=rect.r_y0 / scale,