docling 2.45.0__py3-none-any.whl → 2.46.0__py3-none-any.whl

docling/backend/docling_parse_v4_backend.py

@@ -22,15 +22,52 @@ _log = logging.getLogger(__name__)
 
 
 class DoclingParseV4PageBackend(PdfPageBackend):
-    def __init__(self, parsed_page: SegmentedPdfPage, page_obj: PdfPage):
+    def __init__(
+        self,
+        *,
+        dp_doc: PdfDocument,
+        page_obj: PdfPage,
+        page_no: int,
+        create_words: bool = True,
+        create_textlines: bool = True,
+    ):
         self._ppage = page_obj
-        self._dpage = parsed_page
-        self.valid = parsed_page is not None
+        self._dp_doc = dp_doc
+        self._page_no = page_no
+        self._create_words = create_words
+        self._create_textlines = create_textlines
+
+        self._dpage: Optional[SegmentedPdfPage] = None
+        self._unloaded = False
+        self.valid = (self._ppage is not None) and (self._dp_doc is not None)
+
+    def _ensure_parsed(self) -> None:
+        if self._dpage is not None:
+            return
+
+        seg_page = self._dp_doc.get_page(
+            self._page_no + 1,
+            create_words=self._create_words,
+            create_textlines=self._create_textlines,
+        )
+
+        # In Docling, all TextCell instances are expected with top-left origin.
+        [
+            tc.to_top_left_origin(seg_page.dimension.height)
+            for tc in seg_page.textline_cells
+        ]
+        [tc.to_top_left_origin(seg_page.dimension.height) for tc in seg_page.char_cells]
+        [tc.to_top_left_origin(seg_page.dimension.height) for tc in seg_page.word_cells]
+
+        self._dpage = seg_page
 
     def is_valid(self) -> bool:
         return self.valid
 
     def get_text_in_rect(self, bbox: BoundingBox) -> str:
+        self._ensure_parsed()
+        assert self._dpage is not None
+
         # Find intersecting cells on the page
         text_piece = ""
         page_size = self.get_size()
@@ -56,12 +93,19 @@ class DoclingParseV4PageBackend(PdfPageBackend):
         return text_piece
 
     def get_segmented_page(self) -> Optional[SegmentedPdfPage]:
+        self._ensure_parsed()
         return self._dpage
 
     def get_text_cells(self) -> Iterable[TextCell]:
+        self._ensure_parsed()
+        assert self._dpage is not None
+
         return self._dpage.textline_cells
 
     def get_bitmap_rects(self, scale: float = 1) -> Iterable[BoundingBox]:
+        self._ensure_parsed()
+        assert self._dpage is not None
+
         AREA_THRESHOLD = 0  # 32 * 32
 
         images = self._dpage.bitmap_resources
@@ -123,8 +167,13 @@ class DoclingParseV4PageBackend(PdfPageBackend):
         # )
 
     def unload(self):
+        if not self._unloaded and self._dp_doc is not None:
+            self._dp_doc.unload_pages((self._page_no + 1, self._page_no + 2))
+            self._unloaded = True
+
         self._ppage = None
         self._dpage = None
+        self._dp_doc = None
 
 
 class DoclingParseV4DocumentBackend(PdfDocumentBackend):
@@ -157,30 +206,15 @@ class DoclingParseV4DocumentBackend(PdfDocumentBackend):
         self, page_no: int, create_words: bool = True, create_textlines: bool = True
     ) -> DoclingParseV4PageBackend:
         with pypdfium2_lock:
-            seg_page = self.dp_doc.get_page(
-                page_no + 1,
-                create_words=create_words,
-                create_textlines=create_textlines,
-            )
-
-            # In Docling, all TextCell instances are expected with top-left origin.
-            [
-                tc.to_top_left_origin(seg_page.dimension.height)
-                for tc in seg_page.textline_cells
-            ]
-            [
-                tc.to_top_left_origin(seg_page.dimension.height)
-                for tc in seg_page.char_cells
-            ]
-            [
-                tc.to_top_left_origin(seg_page.dimension.height)
-                for tc in seg_page.word_cells
-            ]
-
-            return DoclingParseV4PageBackend(
-                seg_page,
-                self._pdoc[page_no],
-            )
+            ppage = self._pdoc[page_no]
+
+        return DoclingParseV4PageBackend(
+            dp_doc=self.dp_doc,
+            page_obj=ppage,
+            page_no=page_no,
+            create_words=create_words,
+            create_textlines=create_textlines,
+        )
 
     def is_valid(self) -> bool:
         return self.page_count() > 0

docling/backend/html_backend.py

@@ -38,6 +38,7 @@ _BLOCK_TAGS: Final = {
     "address",
     "details",
     "figure",
+    "footer",
     "h1",
     "h2",
     "h3",
@@ -639,10 +640,12 @@ class HTMLDocumentBackend(DeclarativeDocumentBackend):
                         hyperlink=annotated_text.hyperlink,
                     )
 
-        elif tag_name == "details":
-            # handle details and its content.
+        elif tag_name in {"details", "footer"}:
+            if tag_name == "footer":
+                current_layer = self.content_layer
+                self.content_layer = ContentLayer.FURNITURE
             self.parents[self.level + 1] = doc.add_group(
-                name="details",
+                name=tag_name,
                 label=GroupLabel.SECTION,
                 parent=self.parents[self.level],
                 content_layer=self.content_layer,
@@ -651,6 +654,8 @@ class HTMLDocumentBackend(DeclarativeDocumentBackend):
             self._walk(tag, doc)
             self.parents[self.level + 1] = None
             self.level -= 1
+            if tag_name == "footer":
+                self.content_layer = current_layer
 
     def _emit_image(self, img_tag: Tag, doc: DoclingDocument) -> None:
         figure = img_tag.find_parent("figure")
@@ -686,7 +691,6 @@ class HTMLDocumentBackend(DeclarativeDocumentBackend):
             text_clean = HTMLDocumentBackend._clean_unicode(
                 caption_anno_text.text.strip()
             )
-            print(caption_anno_text)
             caption_item = doc.add_text(
                 label=DocItemLabel.CAPTION,
                 text=text_clean,

docling/datamodel/pipeline_options.py

@@ -323,9 +323,7 @@ class PdfPipelineOptions(PaginatedPipelineOptions):
         ),
     )
 
-    generate_parsed_pages: Literal[True] = (
-        True  # Always True since parsed_page is now mandatory
-    )
+    generate_parsed_pages: bool = False
 
 
 class ProcessingPipeline(str, Enum):

docling/models/code_formula_model.py

@@ -1,5 +1,4 @@
 import re
-from collections import Counter
 from collections.abc import Iterable
 from pathlib import Path
 from typing import List, Literal, Optional, Tuple, Union
@@ -13,10 +12,11 @@ from docling_core.types.doc import (
     TextItem,
 )
 from docling_core.types.doc.labels import CodeLanguageLabel
-from PIL import Image, ImageOps
+from PIL import Image
 from pydantic import BaseModel
+from transformers import AutoModelForImageTextToText, AutoProcessor
 
-from docling.datamodel.accelerator_options import AcceleratorOptions
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 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
@@ -65,9 +65,9 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
         Processes the given batch of elements and enriches them with predictions.
     """
 
-    _model_repo_folder = "ds4sd--CodeFormula"
+    _model_repo_folder = "ds4sd--CodeFormulaV2"
     elements_batch_size = 5
-    images_scale = 1.66  # = 120 dpi, aligned with training data resolution
+    images_scale = 1.67  # = 120 dpi, aligned with training data resolution
     expansion_factor = 0.18
 
     def __init__(
@@ -95,10 +95,9 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
         self.options = options
 
         if self.enabled:
-            device = decide_device(accelerator_options.device)
-
-            from docling_ibm_models.code_formula_model.code_formula_predictor import (
-                CodeFormulaPredictor,
+            self.device = decide_device(
+                accelerator_options.device,
+                supported_devices=[AcceleratorDevice.CPU, AcceleratorDevice.CUDA],
             )
 
             if artifacts_path is None:
@@ -106,11 +105,14 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
             else:
                 artifacts_path = artifacts_path / self._model_repo_folder
 
-            self.code_formula_model = CodeFormulaPredictor(
-                artifacts_path=str(artifacts_path),
-                device=device,
-                num_threads=accelerator_options.num_threads,
+            self._processor = AutoProcessor.from_pretrained(
+                artifacts_path,
+            )
+            self._model_max_length = self._processor.tokenizer.model_max_length
+            self._model = AutoModelForImageTextToText.from_pretrained(
+                artifacts_path, device_map=self.device
             )
+            self._model.eval()
 
     @staticmethod
     def download_models(
@@ -119,8 +121,8 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
         progress: bool = False,
     ) -> Path:
         return download_hf_model(
-            repo_id="ds4sd/CodeFormula",
-            revision="v1.0.2",
+            repo_id="ds4sd/CodeFormulaV2",
+            revision="main",
             local_dir=local_dir,
             force=force,
             progress=progress,
@@ -172,7 +174,7 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
                 - The second element is the extracted language if a match is found;
                 otherwise, `None`.
         """
-        pattern = r"^<_([^_>]+)_>\s(.*)"
+        pattern = r"^<_([^_>]+)_>\s*(.*)"
         match = re.match(pattern, input_string, flags=re.DOTALL)
         if match:
             language = str(match.group(1))  # the captured programming language
@@ -203,81 +205,74 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
         except ValueError:
             return CodeLanguageLabel.UNKNOWN
 
-    def _get_most_frequent_edge_color(self, pil_img: Image.Image):
+    def _get_prompt(self, label: str) -> str:
         """
-        Compute the most frequent color along the outer edges of a PIL image.
+        Constructs the prompt for the model based on the input label.
 
         Parameters
         ----------
-            pil_img : Image.Image
-                A PIL Image in any mode (L, RGB, RGBA, etc.).
+        label : str
+            The type of input, either 'code' or 'formula'.
 
         Returns
         -------
-            (int) or (tuple): The most common edge color as a scalar (for grayscale) or
-                tuple (for RGB/RGBA).
+        str
+            The constructed prompt including necessary tokens and query.
+
+        Raises
+        ------
+        NotImplementedError
+            If the label is not 'code' or 'formula'.
         """
-        # Convert to NumPy array for easy pixel access
-        img_np = np.array(pil_img)
+        if label == "code":
+            query = "<code>"
+        elif label == "formula":
+            query = "<formula>"
+        else:
+            raise NotImplementedError("Label must be either code or formula")
 
-        if img_np.ndim == 2:
-            # Grayscale-like image: shape (H, W)
-            # Extract edges: top row, bottom row, left col, right col
-            top = img_np[0, :]  # shape (W,)
-            bottom = img_np[-1, :]  # shape (W,)
-            left = img_np[:, 0]  # shape (H,)
-            right = img_np[:, -1]  # shape (H,)
+        messages = [
+            {
+                "role": "user",
+                "content": [{"type": "image"}, {"type": "text", "text": query}],
+            },
+        ]
 
-            # Concatenate all edges
-            edges = np.concatenate([top, bottom, left, right])
+        prompt = self._processor.apply_chat_template(
+            messages, add_generation_prompt=True
+        )
 
-            # Count frequencies
-            freq = Counter(edges.tolist())
-            most_common_value, _ = freq.most_common(1)[0]
-            return int(most_common_value)  # single channel color
+        return prompt
 
-        else:
-            # Color image: shape (H, W, C)
-            top = img_np[0, :, :]  # shape (W, C)
-            bottom = img_np[-1, :, :]  # shape (W, C)
-            left = img_np[:, 0, :]  # shape (H, C)
-            right = img_np[:, -1, :]  # shape (H, C)
-
-            # Concatenate edges along first axis
-            edges = np.concatenate([top, bottom, left, right], axis=0)
-
-            # Convert each color to a tuple for counting
-            edges_as_tuples = [tuple(pixel) for pixel in edges]
-            freq = Counter(edges_as_tuples)
-            most_common_value, _ = freq.most_common(1)[0]
-            return most_common_value  # e.g. (R, G, B) or (R, G, B, A)
-
-    def _pad_with_most_frequent_edge_color(
-        self, img: Union[Image.Image, np.ndarray], padding: Tuple[int, int, int, int]
-    ):
+    def _post_process(self, texts: list[str]) -> list[str]:
         """
-        Pads an image (PIL or NumPy array) using the most frequent edge color.
+        Processes a list of text strings by truncating at '<end_of_utterance>' and
+        removing a predefined set of unwanted substrings.
 
         Parameters
         ----------
-            img : Union[Image.Image, np.ndarray]
-                The original image.
-            padding : tuple
-                Padding (left, top, right, bottom) in pixels.
+        texts : list[str]
+            A list of strings to be post-processed.
 
         Returns
         -------
-            Image.Image: A new PIL image with the specified padding.
+        list[str]
+            A list of cleaned strings with specified substrings removed and truncated at
+                '<end_of_utterance>' if present.
         """
-        if isinstance(img, np.ndarray):
-            pil_img = Image.fromarray(img)
-        else:
-            pil_img = img
+        to_remove = ["</code>", "</formula>", "<loc_0><loc_0><loc_500><loc_500>"]
 
-        most_freq_color = self._get_most_frequent_edge_color(pil_img)
+        def clean_text(text: str) -> str:
+            idx = text.find("<end_of_utterance>")
+            if idx != -1:
+                text = text[:idx]
 
-        padded_img = ImageOps.expand(pil_img, border=padding, fill=most_freq_color)
-        return padded_img
+            for token in to_remove:
+                if token in text:
+                    text = text.replace(token, "")
+            return text.lstrip()
+
+        return [clean_text(t) for t in texts]
 
     def __call__(
         self,
@@ -308,14 +303,30 @@ class CodeFormulaModel(BaseItemAndImageEnrichmentModel):
         images: List[Union[Image.Image, np.ndarray]] = []
         elements: List[TextItem] = []
         for el in element_batch:
-            assert isinstance(el.item, TextItem)
-            elements.append(el.item)
-            labels.append(el.item.label)
-            images.append(
-                self._pad_with_most_frequent_edge_color(el.image, (20, 10, 20, 10))
-            )
+            elements.append(el.item)  # type: ignore[arg-type]
+            labels.append(el.item.label)  # type: ignore[attr-defined]
+            images.append(el.image)
+
+        prompts = [self._get_prompt(label) for label in labels]
+        inputs = self._processor(
+            text=prompts,
+            images=images,
+            return_tensors="pt",
+        )
+        inputs = inputs.to(self.device)
 
-        outputs = self.code_formula_model.predict(images, labels)
+        gen_kwargs = dict(
+            max_new_tokens=self._model_max_length - inputs.input_ids.shape[1],
+            use_cache=True,
+            do_sample=False,
+        )
+
+        generated_ids = self._model.generate(**inputs, **gen_kwargs)
+
+        outputs = self._processor.batch_decode(
+            generated_ids[:, inputs.input_ids.shape[1] :], skip_special_tokens=False
+        )
+        outputs = self._post_process(outputs)
 
         for item, output in zip(elements, outputs):
             if isinstance(item, CodeItem):

docling/models/tesseract_ocr_cli_model.py

@@ -320,6 +320,8 @@ class TesseractOcrCliModel(BaseOcrModel):
 
 
 def _parse_orientation(df_osd: pd.DataFrame) -> int:
-    orientations = df_osd.loc[df_osd["key"] == "Orientation in degrees"].value.tolist()
-    orientation = parse_tesseract_orientation(orientations[0].strip())
+    # For strictly optimal performance with invariant dataframe format:
+    mask = df_osd["key"].to_numpy() == "Orientation in degrees"
+    orientation_val = df_osd["value"].to_numpy()[mask][0]
+    orientation = parse_tesseract_orientation(orientation_val.strip())
     return orientation

docling/pipeline/base_pipeline.py

@@ -20,7 +20,7 @@ from docling.datamodel.base_models import (
     Page,
 )
 from docling.datamodel.document import ConversionResult, InputDocument
-from docling.datamodel.pipeline_options import PipelineOptions
+from docling.datamodel.pipeline_options import PdfPipelineOptions, PipelineOptions
 from docling.datamodel.settings import settings
 from docling.models.base_model import GenericEnrichmentModel
 from docling.utils.profiling import ProfilingScope, TimeRecorder
@@ -168,6 +168,12 @@ class PaginatedPipeline(BasePipeline):  # TODO this is a bad name.
                         # Cleanup page backends
                         if not self.keep_backend and p._backend is not None:
                             p._backend.unload()
+                        if (
+                            isinstance(self.pipeline_options, PdfPipelineOptions)
+                            and not self.pipeline_options.generate_parsed_pages
+                        ):
+                            del p.parsed_page
+                            p.parsed_page = None
 
                     end_batch_time = time.monotonic()
                     total_elapsed_time += end_batch_time - start_batch_time

docling/pipeline/threaded_standard_pdf_pipeline.py

@@ -565,10 +565,12 @@ class ThreadedStandardPdfPipeline(BasePipeline):
         if not self.keep_images:
             for p in conv_res.pages:
                 p._image_cache = {}
-        if not self.keep_backend:
-            for p in conv_res.pages:
-                if p._backend is not None:
-                    p._backend.unload()
+        for p in conv_res.pages:
+            if not self.keep_backend and p._backend is not None:
+                p._backend.unload()
+            if not self.pipeline_options.generate_parsed_pages:
+                del p.parsed_page
+                p.parsed_page = None
 
     # ---------------------------------------------------------------- assemble
     def _assemble_document(self, conv_res: ConversionResult) -> ConversionResult:

{docling-2.45.0.dist-info → docling-2.46.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling
-Version: 2.45.0
+Version: 2.46.0
 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
@@ -27,7 +27,7 @@ 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,>=2.42.0
-Requires-Dist: docling-parse<5.0.0,>=4.0.0
+Requires-Dist: docling-parse<5.0.0,>=4.2.2
 Requires-Dist: docling-ibm-models<4,>=3.9.0
 Requires-Dist: filetype<2.0.0,>=1.2.0
 Requires-Dist: pypdfium2!=4.30.1,<5.0.0,>=4.30.0

{docling-2.45.0.dist-info → docling-2.46.0.dist-info}/RECORD

@@ -8,8 +8,8 @@ docling/backend/asciidoc_backend.py,sha256=RDNLrPJHxROiM7-NQdZn3DdvAyiPAndbSWcZo
 docling/backend/csv_backend.py,sha256=2g9famYG2W-ID9jEdZPxc6O8QGv1vWQfjN8pL-QMBE0,4536
 docling/backend/docling_parse_backend.py,sha256=9rUo1vPxX6QLzGqF-2B2iEYglZg6YQ3Uea00XrLluTg,7918
 docling/backend/docling_parse_v2_backend.py,sha256=3ckTfke8IICjaImlIzc3TRhG7KDuxDDba0AuCEcjA-M,9500
-docling/backend/docling_parse_v4_backend.py,sha256=qR_WRVq9JGtRioWCw6MnLWgbvXbC6Y1yds7Ol1-E6UQ,6550
-docling/backend/html_backend.py,sha256=jTkpdJ-EKMmkbUfh88DONVG-gENE7m0_cnIhWpWSobI,34523
+docling/backend/docling_parse_v4_backend.py,sha256=MbCMxNGmoW4iuev9tX1Vt4jtIeak2kC9Uac3xQSRxeo,7509
+docling/backend/html_backend.py,sha256=zJH4wkcyftvoA-ixC4MH-xjwl-TGTN9BvZT7Hhla2mc,34701
 docling/backend/md_backend.py,sha256=qCI7SD9hnWWGrkG_drpzQv2Z7DVBG4Tsq3hhTsYV790,22562
 docling/backend/mets_gbs_backend.py,sha256=EA8sY6tbmGiysKGYPPZiNlK-i7Adn8bLTo-7Ym15hTU,12774
 docling/backend/msexcel_backend.py,sha256=cq8MQ2RSh6pqCiVrldjOerSww7dOPTWmCQoCBI57i6w,18579
@@ -38,7 +38,7 @@ docling/datamodel/asr_model_specs.py,sha256=Wg7z3zm_wXIWu122iPVy0RMECsA_JCFHrlFF
 docling/datamodel/base_models.py,sha256=Ifd8PPHs4sW7ScwSqpa-y3rwgPbde_iw13Y2NUCPfU8,11944
 docling/datamodel/document.py,sha256=zsxFYXvo6GtwGNogSDoBB1TFvkm7IOrP_VnqXNqBhJs,17329
 docling/datamodel/layout_model_specs.py,sha256=GSkJ-Z_0PVgwWGi7C7TsxbzRjlrWS9ZrHJjHumv-Z5U,2339
-docling/datamodel/pipeline_options.py,sha256=TaBmCBRjSxyoh79UkpEkPzokLYS8BA2QJam86g9pT5g,10544
+docling/datamodel/pipeline_options.py,sha256=vOLpuVF-d4nmr-L16EmmhGFn25SDsgExCfX5kPiyISg,10470
 docling/datamodel/pipeline_options_asr_model.py,sha256=7X068xl-qpbyPxC7-TwX7Q6tLyZXGT5h1osZ_xLNLM0,1454
 docling/datamodel/pipeline_options_vlm_model.py,sha256=eH-Cj_8aic9FdX4xGlBcf5_R9e152JAL2LhtY8d0rhw,2498
 docling/datamodel/settings.py,sha256=c0MTw6pO5be_BKxHKYl4SaBJAw_qL-aapxp-g5HHj1A,2084
@@ -47,7 +47,7 @@ docling/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/models/api_vlm_model.py,sha256=-zisU32pgDRbychyG6-neB0qweNbPaYnLXwiGT7SEdI,2859
 docling/models/base_model.py,sha256=NNjIapqCruAEAWR-CCdsNgXc2QkwiPYAcaQ_ZYe1W28,2978
 docling/models/base_ocr_model.py,sha256=kT8TylASOpPlY60rIG6VL6_eLVsfg5KvEVnZHzDWtR0,8193
-docling/models/code_formula_model.py,sha256=5uWh-eI-Ejmv3DujKJoKKgJBuvPLokt7AJ_ybt8VHEw,11373
+docling/models/code_formula_model.py,sha256=XRugm4EwifLRc-TrAk-glKlktJP-nAPneKh2EOovkJU,11308
 docling/models/document_picture_classifier.py,sha256=9JvoWeH5uQBC7levjM8zptk7UT-b8EQnD-2EnxTjTT4,6202
 docling/models/easyocr_model.py,sha256=ECPBd-48cCw5s935NsPJO_C_1QuK_yAUGloMM77WqIM,7387
 docling/models/layout_model.py,sha256=Nfbo6keMB4vVjGoZdFMqD9CmZcWh-0bE3LkRjJTDJQ0,9146
@@ -60,7 +60,7 @@ docling/models/picture_description_vlm_model.py,sha256=yfyAFOy8RjxQJrafPMSAMrrpa
 docling/models/rapid_ocr_model.py,sha256=AMdc66s_iWO4p6nQ0LNjQMUYVxrDSxMyLNPpjPYt6N8,5916
 docling/models/readingorder_model.py,sha256=bZoXHaSwUsa8niSmJrbCuy784ixCeBXT-RQBUfgHJ4A,14925
 docling/models/table_structure_model.py,sha256=RFXo73f2q4XuKyaSqbxpznh7JVtlLcT0FsOWl9oZbSg,12518
-docling/models/tesseract_ocr_cli_model.py,sha256=qcM3-n7Z_dm1CGBhVUcNr2XT41iXnU32zk4RqKHBl9I,12775
+docling/models/tesseract_ocr_cli_model.py,sha256=I3Gn28Y-LD8OfvyCElN9fLiNgpo2sT0uMkVt258253s,12881
 docling/models/tesseract_ocr_model.py,sha256=GdI5Cjfi87qcehVbM3wdKRvKkl_F9A4bwTUbjXZCJYA,10745
 docling/models/factories/__init__.py,sha256=x_EM5dDg_A3HBcBYzOoqwmA2AFLtJ1IzYDPX-R1A-Sg,868
 docling/models/factories/base_factory.py,sha256=MfWIljMETi5aaVR-6qLTelW8u1gwDAQsOwg3fu7O4Qc,4028
@@ -75,10 +75,10 @@ docling/models/vlm_models_inline/hf_transformers_model.py,sha256=Rwdr7neDpn5ehtr
 docling/models/vlm_models_inline/mlx_model.py,sha256=YYYmopsITlX17JVS5KhLlb1IQSEVoSECNx_fXLHNpAc,5880
 docling/pipeline/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/pipeline/asr_pipeline.py,sha256=tQkhu9fXdkSuYIL22xzV2YRUlQh-9qktHBbs2qeXhJI,9070
-docling/pipeline/base_pipeline.py,sha256=MOKZtx3jNYotfntgoJHoyb6UsvdvG6bQLyDl9Lxvc1w,9586
+docling/pipeline/base_pipeline.py,sha256=VYVYndifTPSD2GWHKjfi4Y76M5qgt1DiygO-jowKsqM,9919
 docling/pipeline/simple_pipeline.py,sha256=TXZOwR7hZRji462ZTIpte0VJjzbxvNVE8dbLFANDhSU,2253
 docling/pipeline/standard_pdf_pipeline.py,sha256=yFishq4Cu01BiBGHk3Irr7ogcTQKeSC0QZImQVAhIaY,12740
-docling/pipeline/threaded_standard_pdf_pipeline.py,sha256=Rjdq1x2fRHBA0rMHJ6rqqHzxVVzgTEALBBj5d30oOZ8,26018
+docling/pipeline/threaded_standard_pdf_pipeline.py,sha256=miPIyprtzPFYG94n6PmUgK4Nh7rqACYEGkWrlTbrZAc,26133
 docling/pipeline/vlm_pipeline.py,sha256=0lj8tbXNpYF8OLBoLqP2BZfFpTHi40RoHVfvO_Nah4Q,15349
 docling/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/utils/accelerator_utils.py,sha256=DSajLxVx1JEVT0zt5de26llciLNlVfIDfSa2zYCFJzQ,2909
@@ -93,9 +93,9 @@ docling/utils/orientation.py,sha256=jTyLxyT31FlOodZoBMlADHNQK2lAWKYVs5z7pXd_6Cg,
 docling/utils/profiling.py,sha256=YaMGoB9MMZpagF9mb5ndoHj8Lpb9aIdb7El-Pl7IcFs,1753
 docling/utils/utils.py,sha256=kJtIYuzXeOyJHYlxmLAo7dGM5rEsDa1i84qEsUj1nio,1908
 docling/utils/visualization.py,sha256=tY2ylE2aiQKkmzlSLnFW-HTfFyqUUMguW18ldd1PLfo,2868
-docling-2.45.0.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
-docling-2.45.0.dist-info/METADATA,sha256=-iB6xJ4H7DIStzPn-ruYcBa_Tq45Ijk52zfoM_6FkCE,10459
-docling-2.45.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-docling-2.45.0.dist-info/entry_points.txt,sha256=hzVlbeE0aMSTQ9S0-NTYN0Hmgsn6qL_EA2qX4UbkAuY,149
-docling-2.45.0.dist-info/top_level.txt,sha256=vkIywP-USjFyYo1AIRQbWQQaL3xB5jf8vkCYdTIfNic,8
-docling-2.45.0.dist-info/RECORD,,
+docling-2.46.0.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
+docling-2.46.0.dist-info/METADATA,sha256=fm7KVaUwGryyuRk7R_AkNSHo1BogY8-ra9gpCWXbnCA,10459
+docling-2.46.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+docling-2.46.0.dist-info/entry_points.txt,sha256=hzVlbeE0aMSTQ9S0-NTYN0Hmgsn6qL_EA2qX4UbkAuY,149
+docling-2.46.0.dist-info/top_level.txt,sha256=vkIywP-USjFyYo1AIRQbWQQaL3xB5jf8vkCYdTIfNic,8
+docling-2.46.0.dist-info/RECORD,,