docling 2.39.0__tar.gz → 2.40.0__tar.gz

{docling-2.39.0 → docling-2.40.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling
-Version: 2.39.0
+Version: 2.40.0
 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
@@ -27,8 +27,8 @@ 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-ibm-models<4.0.0,>=3.4.4
 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
 Requires-Dist: pypdfium2<5.0.0,>=4.30.0
 Requires-Dist: pydantic-settings<3.0.0,>=2.3.0
@@ -57,7 +57,7 @@ Requires-Dist: ocrmac<2.0.0,>=1.0.0; sys_platform == "darwin" and extra == "ocrm
 Provides-Extra: vlm
 Requires-Dist: transformers<5.0.0,>=4.46.0; extra == "vlm"
 Requires-Dist: accelerate<2.0.0,>=1.2.1; extra == "vlm"
-Requires-Dist: mlx-vlm>=0.1.22; (python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64") and extra == "vlm"
+Requires-Dist: mlx-vlm<0.2,>=0.1.22; (python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64") and extra == "vlm"
 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"

{docling-2.39.0 → docling-2.40.0}/docling/backend/docling_parse_v4_backend.py

@@ -187,7 +187,17 @@ class DoclingParseV4DocumentBackend(PdfDocumentBackend):
 
     def unload(self):
         super().unload()
-        self.dp_doc.unload()
-        with pypdfium2_lock:
-            self._pdoc.close()
-        self._pdoc = None
+        # Unload docling-parse document first
+        if self.dp_doc is not None:
+            self.dp_doc.unload()
+            self.dp_doc = None
+
+        # Then close pypdfium2 document with proper locking
+        if self._pdoc is not None:
+            with pypdfium2_lock:
+                try:
+                    self._pdoc.close()
+                except Exception:
+                    # Ignore cleanup errors
+                    pass
+            self._pdoc = None

{docling-2.39.0 → docling-2.40.0}/docling/backend/msexcel_backend.py

@@ -337,10 +337,17 @@ class MsExcelDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentBacken
         # Collect the data within the bounds
         data = []
         visited_cells: set[tuple[int, int]] = set()
-        for ri in range(start_row, max_row + 1):
-            for rj in range(start_col, max_col + 1):
-                cell = sheet.cell(row=ri + 1, column=rj + 1)  # 1-based indexing
-
+        for ri, row in enumerate(
+            sheet.iter_rows(
+                min_row=start_row + 1,  # start_row is 0-based but iter_rows is 1-based
+                max_row=max_row + 1,
+                min_col=start_col + 1,
+                max_col=max_col + 1,
+                values_only=False,
+            ),
+            start_row,
+        ):
+            for rj, cell in enumerate(row, start_col):
                 # Check if the cell belongs to a merged range
                 row_span = 1
                 col_span = 1
@@ -397,10 +404,16 @@ class MsExcelDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentBacken
         """
         max_row: int = start_row
 
-        while max_row < sheet.max_row - 1:
-            # Get the cell value or check if it is part of a merged cell
-            cell = sheet.cell(row=max_row + 2, column=start_col + 1)
-
+        for ri, (cell,) in enumerate(
+            sheet.iter_rows(
+                min_row=start_row + 2,
+                max_row=sheet.max_row,
+                min_col=start_col + 1,
+                max_col=start_col + 1,
+                values_only=False,
+            ),
+            start_row + 1,
+        ):
             # Check if the cell is part of a merged range
             merged_range = next(
                 (mr for mr in sheet.merged_cells.ranges if cell.coordinate in mr),
@@ -414,7 +427,7 @@ class MsExcelDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentBacken
             if merged_range:
                 max_row = max(max_row, merged_range.max_row - 1)
             else:
-                max_row += 1
+                max_row = ri
 
         return max_row
 
@@ -433,10 +446,16 @@ class MsExcelDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentBacken
         """
         max_col: int = start_col
 
-        while max_col < sheet.max_column - 1:
-            # Get the cell value or check if it is part of a merged cell
-            cell = sheet.cell(row=start_row + 1, column=max_col + 2)
-
+        for rj, (cell,) in enumerate(
+            sheet.iter_cols(
+                min_row=start_row + 1,
+                max_row=start_row + 1,
+                min_col=start_col + 2,
+                max_col=sheet.max_column,
+                values_only=False,
+            ),
+            start_col + 1,
+        ):
             # Check if the cell is part of a merged range
             merged_range = next(
                 (mr for mr in sheet.merged_cells.ranges if cell.coordinate in mr),
@@ -450,7 +469,7 @@ class MsExcelDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentBacken
             if merged_range:
                 max_col = max(max_col, merged_range.max_col - 1)
             else:
-                max_col += 1
+                max_col = rj
 
         return max_col
 

{docling-2.39.0 → docling-2.40.0}/docling/datamodel/pipeline_options.py

@@ -1,4 +1,5 @@
 import logging
+from datetime import datetime
 from enum import Enum
 from pathlib import Path
 from typing import Any, ClassVar, Dict, List, Literal, Optional, Union
@@ -265,6 +266,12 @@ class VlmPipelineOptions(PaginatedPipelineOptions):
     )
 
 
+class LayoutOptions(BaseModel):
+    """Options for layout processing."""
+
+    create_orphan_clusters: bool = True  # Whether to create clusters for orphaned cells
+
+
 class AsrPipelineOptions(PipelineOptions):
     asr_options: Union[InlineAsrOptions] = asr_model_specs.WHISPER_TINY
     artifacts_path: Optional[Union[Path, str]] = None
@@ -289,6 +296,7 @@ class PdfPipelineOptions(PaginatedPipelineOptions):
     picture_description_options: PictureDescriptionBaseOptions = (
         smolvlm_picture_description
     )
+    layout_options: LayoutOptions = LayoutOptions()
 
     images_scale: float = 1.0
     generate_page_images: bool = False

{docling-2.39.0 → docling-2.40.0}/docling/models/base_ocr_model.py

@@ -3,14 +3,13 @@ import logging
 from abc import abstractmethod
 from collections.abc import Iterable
 from pathlib import Path
-from typing import List, Optional, Type
+from typing import TYPE_CHECKING, List, Optional, Type
 
 import numpy as np
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import TextCell
 from PIL import Image, ImageDraw
 from rtree import index
-from scipy.ndimage import binary_dilation, find_objects, label
 
 from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import Page
@@ -31,11 +30,16 @@ class BaseOcrModel(BasePageModel, BaseModelWithOptions):
         options: OcrOptions,
         accelerator_options: AcceleratorOptions,
     ):
+        # Make sure any delay/error from import occurs on ocr model init and not first use
+        from scipy.ndimage import binary_dilation, find_objects, label
+
         self.enabled = enabled
         self.options = options
 
     # Computes the optimum amount and coordinates of rectangles to OCR on a given page
     def get_ocr_rects(self, page: Page) -> List[BoundingBox]:
+        from scipy.ndimage import binary_dilation, find_objects, label
+
         BITMAP_COVERAGE_TRESHOLD = 0.75
         assert page.size is not None
 

{docling-2.39.0 → docling-2.40.0}/docling/models/layout_model.py

@@ -7,12 +7,12 @@ from typing import Optional
 
 import numpy as np
 from docling_core.types.doc import DocItemLabel
-from docling_ibm_models.layoutmodel.layout_predictor import LayoutPredictor
 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.pipeline_options import LayoutOptions
 from docling.datamodel.settings import settings
 from docling.models.base_model import BasePageModel
 from docling.models.utils.hf_model_download import download_hf_model
@@ -49,8 +49,15 @@ class LayoutModel(BasePageModel):
     CONTAINER_LABELS = [DocItemLabel.FORM, DocItemLabel.KEY_VALUE_REGION]
 
     def __init__(
-        self, artifacts_path: Optional[Path], accelerator_options: AcceleratorOptions
+        self,
+        artifacts_path: Optional[Path],
+        accelerator_options: AcceleratorOptions,
+        options: LayoutOptions,
     ):
+        from docling_ibm_models.layoutmodel.layout_predictor import LayoutPredictor
+
+        self.options = options
+
         device = decide_device(accelerator_options.device)
 
         if artifacts_path is None:
@@ -176,7 +183,7 @@ class LayoutModel(BasePageModel):
                     # Apply postprocessing
 
                     processed_clusters, processed_cells = LayoutPostprocessor(
-                        page, clusters
+                        page, clusters, self.options
                     ).postprocess()
                     # Note: LayoutPostprocessor updates page.cells and page.parsed_page internally
 

{docling-2.39.0 → docling-2.40.0}/docling/models/picture_description_vlm_model.py

@@ -1,3 +1,4 @@
+import threading
 from collections.abc import Iterable
 from pathlib import Path
 from typing import Optional, Type, Union
@@ -15,6 +16,9 @@ from docling.models.utils.hf_model_download import (
 )
 from docling.utils.accelerator_utils import decide_device
 
+# Global lock for model initialization to prevent threading issues
+_model_init_lock = threading.Lock()
+
 
 class PictureDescriptionVlmModel(
     PictureDescriptionBaseModel, HuggingFaceModelDownloadMixin
@@ -57,17 +61,18 @@ class PictureDescriptionVlmModel(
                 )
 
             # Initialize processor and model
-            self.processor = AutoProcessor.from_pretrained(artifacts_path)
-            self.model = AutoModelForVision2Seq.from_pretrained(
-                artifacts_path,
-                torch_dtype=torch.bfloat16,
-                _attn_implementation=(
-                    "flash_attention_2"
-                    if self.device.startswith("cuda")
-                    and accelerator_options.cuda_use_flash_attention2
-                    else "eager"
-                ),
-            ).to(self.device)
+            with _model_init_lock:
+                self.processor = AutoProcessor.from_pretrained(artifacts_path)
+                self.model = AutoModelForVision2Seq.from_pretrained(
+                    artifacts_path,
+                    torch_dtype=torch.bfloat16,
+                    _attn_implementation=(
+                        "flash_attention_2"
+                        if self.device.startswith("cuda")
+                        and accelerator_options.cuda_use_flash_attention2
+                        else "eager"
+                    ),
+                ).to(self.device)
 
             self.provenance = f"{self.options.repo_id}"
 

docling-2.40.0/docling/models/plugins/defaults.py

@@ -0,0 +1,28 @@
+def ocr_engines():
+    from docling.models.easyocr_model import EasyOcrModel
+    from docling.models.ocr_mac_model import OcrMacModel
+    from docling.models.rapid_ocr_model import RapidOcrModel
+    from docling.models.tesseract_ocr_cli_model import TesseractOcrCliModel
+    from docling.models.tesseract_ocr_model import TesseractOcrModel
+
+    return {
+        "ocr_engines": [
+            EasyOcrModel,
+            OcrMacModel,
+            RapidOcrModel,
+            TesseractOcrModel,
+            TesseractOcrCliModel,
+        ]
+    }
+
+
+def picture_description():
+    from docling.models.picture_description_api_model import PictureDescriptionApiModel
+    from docling.models.picture_description_vlm_model import PictureDescriptionVlmModel
+
+    return {
+        "picture_description": [
+            PictureDescriptionVlmModel,
+            PictureDescriptionApiModel,
+        ]
+    }

{docling-2.39.0 → docling-2.40.0}/docling/models/readingorder_model.py

@@ -12,6 +12,9 @@ from docling_core.types.doc import (
     TableData,
 )
 from docling_core.types.doc.document import ContentLayer
+from docling_ibm_models.list_item_normalizer.list_marker_processor import (
+    ListItemMarkerProcessor,
+)
 from docling_ibm_models.reading_order.reading_order_rb import (
     PageElement as ReadingOrderPageElement,
     ReadingOrderPredictor,
@@ -40,6 +43,7 @@ class ReadingOrderModel:
     def __init__(self, options: ReadingOrderOptions):
         self.options = options
         self.ro_model = ReadingOrderPredictor()
+        self.list_item_processor = ListItemMarkerProcessor()
 
     def _assembled_to_readingorder_elements(
         self, conv_res: ConversionResult
@@ -92,7 +96,8 @@ class ReadingOrderModel:
             )
             if c_label == DocItemLabel.LIST_ITEM:
                 # TODO: Infer if this is a numbered or a bullet list item
-                doc.add_list_item(parent=doc_item, text=c_text, prov=c_prov)
+                l_item = doc.add_list_item(parent=doc_item, text=c_text, prov=c_prov)
+                self.list_item_processor.process_list_item(l_item)
             elif c_label == DocItemLabel.SECTION_HEADER:
                 doc.add_heading(parent=doc_item, text=c_text, prov=c_prov)
             else:
@@ -301,6 +306,8 @@ class ReadingOrderModel:
             new_item = out_doc.add_list_item(
                 text=cap_text, enumerated=False, prov=prov, parent=current_list
             )
+            self.list_item_processor.process_list_item(new_item)
+
         elif label == DocItemLabel.SECTION_HEADER:
             current_list = None
 

{docling-2.39.0 → docling-2.40.0}/docling/models/table_structure_model.py

@@ -10,7 +10,6 @@ from docling_core.types.doc.page import (
     BoundingRectangle,
     TextCellUnit,
 )
-from docling_ibm_models.tableformer.data_management.tf_predictor import TFPredictor
 from PIL import ImageDraw
 
 from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
@@ -70,6 +69,9 @@ class TableStructureModel(BasePageModel):
 
             # Third Party
             import docling_ibm_models.tableformer.common as c
+            from docling_ibm_models.tableformer.data_management.tf_predictor import (
+                TFPredictor,
+            )
 
             device = decide_device(accelerator_options.device)
 

{docling-2.39.0 → docling-2.40.0}/docling/models/tesseract_ocr_model.py

@@ -144,7 +144,10 @@ class TesseractOcrModel(BaseOcrModel):
 
                         local_reader = self.reader
                         self.osd_reader.SetImage(high_res_image)
+
+                        doc_orientation = 0
                         osd = self.osd_reader.DetectOrientationScript()
+
                         # No text, or Orientation and Script detection failure
                         if osd is None:
                             _log.error(
@@ -158,11 +161,14 @@ class TesseractOcrModel(BaseOcrModel):
                             # to OCR in the hope OCR will succeed while OSD failed
                             if self._is_auto:
                                 continue
-                        doc_orientation = parse_tesseract_orientation(osd["orient_deg"])
-                        if doc_orientation != 0:
-                            high_res_image = high_res_image.rotate(
-                                -doc_orientation, expand=True
+                        else:
+                            doc_orientation = parse_tesseract_orientation(
+                                osd["orient_deg"]
                             )
+                            if doc_orientation != 0:
+                                high_res_image = high_res_image.rotate(
+                                    -doc_orientation, expand=True
+                                )
                         if self._is_auto:
                             script = osd["script_name"]
                             script = map_tesseract_script(script)

{docling-2.39.0 → docling-2.40.0}/docling/pipeline/standard_pdf_pipeline.py

@@ -80,6 +80,7 @@ class StandardPdfPipeline(PaginatedPipeline):
             LayoutModel(
                 artifacts_path=artifacts_path,
                 accelerator_options=pipeline_options.accelerator_options,
+                options=pipeline_options.layout_options,
             ),
             # Table structure model
             TableStructureModel(

{docling-2.39.0 → docling-2.40.0}/docling/utils/accelerator_utils.py

@@ -1,8 +1,6 @@
 import logging
 from typing import List, Optional
 
-import torch
-
 from docling.datamodel.accelerator_options import AcceleratorDevice
 
 _log = logging.getLogger(__name__)
@@ -18,6 +16,8 @@ def decide_device(
     1. AUTO: Check for the best available device on the system.
     2. User-defined: Check if the device actually exists, otherwise fall-back to CPU
     """
+    import torch
+
     device = "cpu"
 
     has_cuda = torch.backends.cuda.is_built() and torch.cuda.is_available()

{docling-2.39.0 → docling-2.40.0}/docling/utils/layout_postprocessor.py

@@ -9,6 +9,7 @@ from docling_core.types.doc.page import TextCell
 from rtree import index
 
 from docling.datamodel.base_models import BoundingBox, Cluster, Page
+from docling.datamodel.pipeline_options import LayoutOptions
 
 _log = logging.getLogger(__name__)
 
@@ -194,12 +195,16 @@ class LayoutPostprocessor:
         DocItemLabel.TITLE: DocItemLabel.SECTION_HEADER,
     }
 
-    def __init__(self, page: Page, clusters: List[Cluster]) -> None:
+    def __init__(
+        self, page: Page, clusters: List[Cluster], options: LayoutOptions
+    ) -> None:
         """Initialize processor with page and clusters."""
+
         self.cells = page.cells
         self.page = page
         self.page_size = page.size
         self.all_clusters = clusters
+        self.options = options
         self.regular_clusters = [
             c for c in clusters if c.label not in self.SPECIAL_TYPES
         ]
@@ -267,7 +272,7 @@ class LayoutPostprocessor:
 
         # Handle orphaned cells
         unassigned = self._find_unassigned_cells(clusters)
-        if unassigned:
+        if unassigned and self.options.create_orphan_clusters:
             next_id = max((c.id for c in self.all_clusters), default=0) + 1
             orphan_clusters = []
             for i, cell in enumerate(unassigned):

{docling-2.39.0 → docling-2.40.0}/docling.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling
-Version: 2.39.0
+Version: 2.40.0
 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
@@ -27,8 +27,8 @@ 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-ibm-models<4.0.0,>=3.4.4
 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
 Requires-Dist: pypdfium2<5.0.0,>=4.30.0
 Requires-Dist: pydantic-settings<3.0.0,>=2.3.0
@@ -57,7 +57,7 @@ Requires-Dist: ocrmac<2.0.0,>=1.0.0; sys_platform == "darwin" and extra == "ocrm
 Provides-Extra: vlm
 Requires-Dist: transformers<5.0.0,>=4.46.0; extra == "vlm"
 Requires-Dist: accelerate<2.0.0,>=1.2.1; extra == "vlm"
-Requires-Dist: mlx-vlm>=0.1.22; (python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64") and extra == "vlm"
+Requires-Dist: mlx-vlm<0.2,>=0.1.22; (python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64") and extra == "vlm"
 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"

{docling-2.39.0 → docling-2.40.0}/docling.egg-info/requires.txt

@@ -1,7 +1,7 @@
 pydantic<3.0.0,>=2.0.0
 docling-core[chunking]<3.0.0,>=2.39.0
-docling-ibm-models<4.0.0,>=3.4.4
 docling-parse<5.0.0,>=4.0.0
+docling-ibm-models<4,>=3.6.0
 filetype<2.0.0,>=1.2.0
 pypdfium2<5.0.0,>=4.30.0
 pydantic-settings<3.0.0,>=2.3.0
@@ -46,4 +46,4 @@ transformers<5.0.0,>=4.46.0
 accelerate<2.0.0,>=1.2.1
 
 [vlm:python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"]
-mlx-vlm>=0.1.22
+mlx-vlm<0.2,>=0.1.22

{docling-2.39.0 → docling-2.40.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "docling"
-version = "2.39.0"  # DO NOT EDIT, updated automatically
+version = "2.40.0"  # DO NOT EDIT, updated automatically
 description = "SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications."
 license = "MIT"
 keywords = [
@@ -45,8 +45,8 @@ requires-python = '>=3.9,<4.0'
 dependencies = [
   'pydantic (>=2.0.0,<3.0.0)',
   'docling-core[chunking] (>=2.39.0,<3.0.0)',
-  'docling-ibm-models (>=3.4.4,<4.0.0)',
   'docling-parse (>=4.0.0,<5.0.0)',
+  'docling-ibm-models (>=3.6.0,<4)',
   'filetype (>=1.2.0,<2.0.0)',
   'pypdfium2 (>=4.30.0,<5.0.0)',
   'pydantic-settings (>=2.3.0,<3.0.0)',
@@ -91,7 +91,7 @@ ocrmac = ['ocrmac (>=1.0.0,<2.0.0) ; sys_platform == "darwin"']
 vlm = [
   'transformers (>=4.46.0,<5.0.0)',
   'accelerate (>=1.2.1,<2.0.0)',
-  'mlx-vlm >=0.1.22 ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
+  'mlx-vlm (>=0.1.22,<0.2) ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
 ]
 rapidocr = [
   'rapidocr-onnxruntime (>=1.4.0,<2.0.0) ; python_version < "3.13"',

{docling-2.39.0 → docling-2.40.0}/tests/test_backend_docling_parse_v4.py

@@ -46,6 +46,12 @@ def test_text_cell_counts():
                 )
             last_cell_count = len(cells)
 
+            # Clean up page backend after each iteration
+            page_backend.unload()
+
+    # Explicitly clean up document backend to prevent race conditions in CI
+    doc_backend.unload()
+
 
 def test_get_text_from_rect(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
@@ -59,6 +65,10 @@ def test_get_text_from_rect(test_doc_path):
 
     assert textpiece.strip() == ref
 
+    # Explicitly clean up resources
+    page_backend.unload()
+    doc_backend.unload()
+
 
 def test_crop_page_image(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
@@ -70,7 +80,14 @@ def test_crop_page_image(test_doc_path):
     )
     # im.show()
 
+    # Explicitly clean up resources
+    page_backend.unload()
+    doc_backend.unload()
+
 
 def test_num_pages(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
     doc_backend.page_count() == 9
+
+    # Explicitly clean up resources to prevent race conditions in CI
+    doc_backend.unload()

docling-2.39.0/docling/models/plugins/defaults.py

@@ -1,28 +0,0 @@
-from docling.models.easyocr_model import EasyOcrModel
-from docling.models.ocr_mac_model import OcrMacModel
-from docling.models.picture_description_api_model import PictureDescriptionApiModel
-from docling.models.picture_description_vlm_model import PictureDescriptionVlmModel
-from docling.models.rapid_ocr_model import RapidOcrModel
-from docling.models.tesseract_ocr_cli_model import TesseractOcrCliModel
-from docling.models.tesseract_ocr_model import TesseractOcrModel
-
-
-def ocr_engines():
-    return {
-        "ocr_engines": [
-            EasyOcrModel,
-            OcrMacModel,
-            RapidOcrModel,
-            TesseractOcrModel,
-            TesseractOcrCliModel,
-        ]
-    }
-
-
-def picture_description():
-    return {
-        "picture_description": [
-            PictureDescriptionVlmModel,
-            PictureDescriptionApiModel,
-        ]
-    }