docling 2.43.0__py3-none-any.whl → 2.45.0__py3-none-any.whl

docling/backend/mets_gbs_backend.py

@@ -0,0 +1,399 @@
+"""Backend for GBS Google Books schema."""
+
+import logging
+import tarfile
+from collections.abc import Iterable
+from dataclasses import dataclass
+from enum import Enum
+from io import BytesIO
+from pathlib import Path
+from typing import TYPE_CHECKING, Dict, List, Optional, Set, Tuple, Union
+
+from docling_core.types.doc import BoundingBox, CoordOrigin, Size
+from docling_core.types.doc.page import (
+    BoundingRectangle,
+    PdfPageBoundaryType,
+    PdfPageGeometry,
+    SegmentedPdfPage,
+    TextCell,
+)
+from lxml import etree
+from PIL import Image
+from PIL.Image import Image as PILImage
+
+from docling.backend.abstract_backend import PaginatedDocumentBackend
+from docling.backend.pdf_backend import PdfDocumentBackend, PdfPageBackend
+from docling.datamodel.base_models import InputFormat
+
+if TYPE_CHECKING:
+    from docling.datamodel.document import InputDocument
+
+_log = logging.getLogger(__name__)
+
+
+def _get_pdf_page_geometry(
+    size: Size,
+) -> PdfPageGeometry:
+    boundary_type: PdfPageBoundaryType = PdfPageBoundaryType.CROP_BOX
+
+    bbox_tuple = (0, 0, size.width, size.height)
+    bbox = BoundingBox.from_tuple(bbox_tuple, CoordOrigin.TOPLEFT)
+
+    return PdfPageGeometry(
+        angle=0.0,
+        rect=BoundingRectangle.from_bounding_box(bbox),
+        boundary_type=boundary_type,
+        art_bbox=bbox,
+        bleed_bbox=bbox,
+        crop_bbox=bbox,
+        media_bbox=bbox,
+        trim_bbox=bbox,
+    )
+
+
+class MetsGbsPageBackend(PdfPageBackend):
+    def __init__(self, parsed_page: SegmentedPdfPage, page_im: PILImage):
+        self._im = page_im
+        self._dpage = parsed_page
+        self.valid = parsed_page is not None
+
+    def is_valid(self) -> bool:
+        return self.valid
+
+    def get_text_in_rect(self, bbox: BoundingBox) -> str:
+        # Find intersecting cells on the page
+        text_piece = ""
+        page_size = self.get_size()
+
+        scale = (
+            1  # FIX - Replace with param in get_text_in_rect across backends (optional)
+        )
+
+        for i, cell in enumerate(self._dpage.textline_cells):
+            cell_bbox = (
+                cell.rect.to_bounding_box()
+                .to_top_left_origin(page_height=page_size.height)
+                .scaled(scale)
+            )
+
+            overlap_frac = cell_bbox.intersection_over_self(bbox)
+
+            if overlap_frac > 0.5:
+                if len(text_piece) > 0:
+                    text_piece += " "
+                text_piece += cell.text
+
+        return text_piece
+
+    def get_segmented_page(self) -> Optional[SegmentedPdfPage]:
+        return self._dpage
+
+    def get_text_cells(self) -> Iterable[TextCell]:
+        return self._dpage.textline_cells
+
+    def get_bitmap_rects(self, scale: float = 1) -> Iterable[BoundingBox]:
+        AREA_THRESHOLD = 0  # 32 * 32
+
+        images = self._dpage.bitmap_resources
+
+        for img in images:
+            cropbox = img.rect.to_bounding_box().to_top_left_origin(
+                self.get_size().height
+            )
+
+            if cropbox.area() > AREA_THRESHOLD:
+                cropbox = cropbox.scaled(scale=scale)
+
+                yield cropbox
+
+    def get_page_image(
+        self, scale: float = 1, cropbox: Optional[BoundingBox] = None
+    ) -> Image.Image:
+        page_size = self.get_size()
+        assert (
+            page_size.width == self._im.size[0] and page_size.height == self._im.size[1]
+        )
+
+        if not cropbox:
+            cropbox = BoundingBox(
+                l=0,
+                r=page_size.width,
+                t=0,
+                b=page_size.height,
+                coord_origin=CoordOrigin.TOPLEFT,
+            )
+
+        image = self._im.resize(
+            size=(round(page_size.width * scale), round(page_size.height * scale))
+        ).crop(cropbox.scaled(scale=scale).as_tuple())
+        return image
+
+    def get_size(self) -> Size:
+        return Size(
+            width=self._dpage.dimension.width, height=self._dpage.dimension.height
+        )
+
+    def unload(self) -> None:
+        if hasattr(self, "_im"):
+            delattr(self, "_im")
+        if hasattr(self, "_dpage"):
+            delattr(self, "_dpage")
+
+
+class _UseType(str, Enum):
+    IMAGE = "image"
+    OCR = "OCR"
+    COORD_OCR = "coordOCR"
+
+
+@dataclass
+class _FileInfo:
+    file_id: str
+    mimetype: str
+    path: str
+    use: _UseType
+
+
+@dataclass
+class _PageFiles:
+    image: Optional[_FileInfo] = None
+    ocr: Optional[_FileInfo] = None
+    coordOCR: Optional[_FileInfo] = None
+
+
+def _extract_rect(title_str: str) -> Optional[BoundingRectangle]:
+    """
+    Extracts bbox from title string like 'bbox 279 177 306 214;x_wconf 97'
+    """
+    parts = title_str.split(";")
+    for part in parts:
+        part = part.strip()
+        if part.startswith("bbox "):
+            try:
+                coords = part.split()[1:]
+                rect = BoundingRectangle.from_bounding_box(
+                    bbox=BoundingBox.from_tuple(
+                        tuple(map(int, coords)), origin=CoordOrigin.TOPLEFT
+                    )
+                )
+                return rect
+            except Exception:
+                return None
+    return None
+
+
+def _extract_confidence(title_str) -> float:
+    """Extracts x_wconf (OCR confidence) value from title string."""
+    for part in title_str.split(";"):
+        part = part.strip()
+        if part.startswith("x_wconf"):
+            try:
+                return float(part.split()[1]) / 100.0
+            except Exception:
+                return 1
+    return 1
+
+
+class MetsGbsDocumentBackend(PdfDocumentBackend):
+    def __init__(self, in_doc: "InputDocument", path_or_stream: Union[BytesIO, Path]):
+        super().__init__(in_doc, path_or_stream)
+
+        self._tar: tarfile.TarFile = (
+            tarfile.open(name=self.path_or_stream, mode="r:gz")
+            if isinstance(self.path_or_stream, Path)
+            else tarfile.open(fileobj=self.path_or_stream, mode="r:gz")
+        )
+        self.root_mets: Optional[etree._Element] = None
+        self.page_map: Dict[int, _PageFiles] = {}
+
+        for member in self._tar.getmembers():
+            if member.name.endswith(".xml"):
+                file = self._tar.extractfile(member)
+                if file is not None:
+                    content = file.read()
+                    self.root_mets = self._validate_mets_xml(content)
+                    if self.root_mets is not None:
+                        break
+
+        if self.root_mets is None:
+            raise RuntimeError(
+                f"METS GBS backend could not load document {self.document_hash}."
+            )
+
+        ns = {
+            "mets": "http://www.loc.gov/METS/",
+            "xlink": "http://www.w3.org/1999/xlink",
+            "xsi": "http://www.w3.org/2001/XMLSchema-instance",
+            "gbs": "http://books.google.com/gbs",
+            "premis": "info:lc/xmlns/premis-v2",
+            "marc": "http://www.loc.gov/MARC21/slim",
+        }
+
+        file_info_by_id: Dict[str, _FileInfo] = {}
+
+        for filegrp in self.root_mets.xpath(".//mets:fileGrp", namespaces=ns):
+            use_raw = filegrp.get("USE")
+            try:
+                use = _UseType(use_raw)
+            except ValueError:
+                continue  # Ignore unknown USE types
+
+            for file_elem in filegrp.xpath("./mets:file", namespaces=ns):
+                file_id = file_elem.get("ID")
+                mimetype = file_elem.get("MIMETYPE")
+                flocat_elem = file_elem.find("mets:FLocat", namespaces=ns)
+                href = (
+                    flocat_elem.get("{http://www.w3.org/1999/xlink}href")
+                    if flocat_elem is not None
+                    else None
+                )
+                if href is None:
+                    continue
+
+                file_info_by_id[file_id] = _FileInfo(
+                    file_id=file_id, mimetype=mimetype, path=href, use=use
+                )
+
+        USE_TO_ATTR = {
+            _UseType.IMAGE: "image",
+            _UseType.OCR: "ocr",
+            _UseType.COORD_OCR: "coordOCR",
+        }
+
+        for div in self.root_mets.xpath('.//mets:div[@TYPE="page"]', namespaces=ns):
+            order_str = div.get("ORDER")
+            if not order_str:
+                continue
+            try:
+                page_no = int(order_str) - 1  # make 0-index pages
+            except ValueError:
+                continue
+
+            page_files = _PageFiles()
+
+            for fptr in div.xpath("./mets:fptr", namespaces=ns):
+                file_id = fptr.get("FILEID")
+                file_info = file_info_by_id.get(file_id)
+
+                if file_info:
+                    attr = USE_TO_ATTR.get(file_info.use)
+                    if attr:
+                        setattr(page_files, attr, file_info)
+
+            self.page_map[page_no] = page_files
+
+    def _validate_mets_xml(self, xml_string) -> Optional[etree._Element]:
+        root: etree._Element = etree.fromstring(xml_string)
+        if (
+            root.tag == "{http://www.loc.gov/METS/}mets"
+            and root.get("PROFILE") == "gbs"
+        ):
+            return root
+
+        _log.warning(f"The root element is not <mets:mets> with PROFILE='gbs': {root}")
+        return None
+
+    def _parse_page(self, page_no: int) -> Tuple[SegmentedPdfPage, PILImage]:
+        # TODO: use better fallbacks...
+        image_info = self.page_map[page_no].image
+        assert image_info is not None
+        ocr_info = self.page_map[page_no].coordOCR
+        assert ocr_info is not None
+
+        image_file = self._tar.extractfile(image_info.path)
+        assert image_file is not None
+        buf = BytesIO(image_file.read())
+        im: PILImage = Image.open(buf)
+        ocr_file = self._tar.extractfile(ocr_info.path)
+        assert ocr_file is not None
+        ocr_content = ocr_file.read()
+        parser = etree.HTMLParser()
+        ocr_root: etree._Element = etree.fromstring(ocr_content, parser=parser)
+
+        line_cells: List[TextCell] = []
+        word_cells: List[TextCell] = []
+
+        page_div = ocr_root.xpath("//div[@class='ocr_page']")
+
+        size = Size(width=im.size[0], height=im.size[1])
+        if page_div:
+            title = page_div[0].attrib.get("title", "")
+            rect = _extract_rect(title)
+            if rect:
+                size = Size(width=rect.width, height=rect.height)
+        else:
+            _log.error(f"Could not find ocr_page for page {page_no}")
+
+        im = im.resize(size=(round(size.width), round(size.height)))
+        im = im.convert("RGB")
+
+        # Extract all ocrx_word spans
+        for ix, word in enumerate(ocr_root.xpath("//span[@class='ocrx_word']")):
+            text = "".join(word.itertext()).strip()
+            title = word.attrib.get("title", "")
+            rect = _extract_rect(title)
+            conf = _extract_confidence(title)
+            if rect:
+                word_cells.append(
+                    TextCell(
+                        index=ix,
+                        text=text,
+                        orig=text,
+                        rect=rect,
+                        from_ocr=True,
+                        confidence=conf,
+                    )
+                )
+
+        # Extract all ocr_line spans
+        # line: etree._Element
+        for ix, line in enumerate(ocr_root.xpath("//span[@class='ocr_line']")):
+            text = "".join(line.itertext()).strip()
+            title = line.attrib.get("title", "")
+            rect = _extract_rect(title)
+            conf = _extract_confidence(title)
+            if rect:
+                line_cells.append(
+                    TextCell(
+                        index=ix,
+                        text=text,
+                        orig=text,
+                        rect=rect,
+                        from_ocr=True,
+                        confidence=conf,
+                    )
+                )
+
+        page = SegmentedPdfPage(
+            dimension=_get_pdf_page_geometry(size),
+            textline_cells=line_cells,
+            char_cells=[],
+            word_cells=word_cells,
+            has_textlines=True,
+            has_words=True,
+            has_chars=False,
+        )
+        return page, im
+
+    def page_count(self) -> int:
+        return len(self.page_map)
+
+    def load_page(self, page_no: int) -> MetsGbsPageBackend:
+        # TODO: is this thread-safe?
+        page, im = self._parse_page(page_no)
+        return MetsGbsPageBackend(parsed_page=page, page_im=im)
+
+    def is_valid(self) -> bool:
+        return self.root_mets is not None and self.page_count() > 0
+
+    @classmethod
+    def supported_formats(cls) -> Set[InputFormat]:
+        return {InputFormat.METS_GBS}
+
+    @classmethod
+    def supports_pagination(cls) -> bool:
+        return True
+
+    def unload(self) -> None:
+        super().unload()
+        self._tar.close()

docling/backend/pdf_backend.py

@@ -84,9 +84,9 @@ class PdfDocumentBackend(PaginatedDocumentBackend):
 
                 buf.seek(0)
                 self.path_or_stream = buf
-            else:
+            elif self.input_format not in self.supported_formats():
                 raise RuntimeError(
-                    f"Incompatible file format {self.input_format} was passed to a PdfDocumentBackend."
+                    f"Incompatible file format {self.input_format} was passed to a PdfDocumentBackend. Valid format are {','.join(self.supported_formats())}."
                 )
 
     @abstractmethod
@@ -99,7 +99,7 @@ class PdfDocumentBackend(PaginatedDocumentBackend):
 
     @classmethod
     def supported_formats(cls) -> Set[InputFormat]:
-        return {InputFormat.PDF}
+        return {InputFormat.PDF, InputFormat.IMAGE}
 
     @classmethod
     def supports_pagination(cls) -> bool:

docling/cli/main.py

@@ -26,6 +26,7 @@ from rich.console import Console
 from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
 from docling.backend.docling_parse_v2_backend import DoclingParseV2DocumentBackend
 from docling.backend.docling_parse_v4_backend import DoclingParseV4DocumentBackend
+from docling.backend.mets_gbs_backend import MetsGbsDocumentBackend
 from docling.backend.pdf_backend import PdfDocumentBackend
 from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
 from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
@@ -262,6 +263,12 @@ def export_documents(
 
         else:
             _log.warning(f"Document {conv_res.input.file} failed to convert.")
+            if _log.isEnabledFor(logging.INFO):
+                for err in conv_res.errors:
+                    _log.info(
+                        f"  [Failure Detail] Component: {err.component_type}, "
+                        f"Module: {err.module_name}, Message: {err.error_message}"
+                    )
             failure_count += 1
 
     _log.info(
@@ -601,9 +608,18 @@ def convert(  # noqa: C901
                 backend=backend,  # pdf_backend
             )
 
+            # METS GBS options
+            mets_gbs_options = pipeline_options.model_copy()
+            mets_gbs_options.do_ocr = False
+            mets_gbs_format_option = PdfFormatOption(
+                pipeline_options=mets_gbs_options,
+                backend=MetsGbsDocumentBackend,
+            )
+
             format_options = {
                 InputFormat.PDF: pdf_format_option,
                 InputFormat.IMAGE: pdf_format_option,
+                InputFormat.METS_GBS: mets_gbs_format_option,
             }
 
         elif pipeline == ProcessingPipeline.VLM:

docling/datamodel/base_models.py

@@ -56,6 +56,7 @@ class InputFormat(str, Enum):
     XLSX = "xlsx"
     XML_USPTO = "xml_uspto"
     XML_JATS = "xml_jats"
+    METS_GBS = "mets_gbs"
     JSON_DOCLING = "json_docling"
     AUDIO = "audio"
 
@@ -81,6 +82,7 @@ FormatToExtensions: Dict[InputFormat, List[str]] = {
     InputFormat.CSV: ["csv"],
     InputFormat.XLSX: ["xlsx", "xlsm"],
     InputFormat.XML_USPTO: ["xml", "txt"],
+    InputFormat.METS_GBS: ["tar.gz"],
     InputFormat.JSON_DOCLING: ["json"],
     InputFormat.AUDIO: ["wav", "mp3"],
 }
@@ -113,6 +115,7 @@ FormatToMimeType: Dict[InputFormat, List[str]] = {
         "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
     ],
     InputFormat.XML_USPTO: ["application/xml", "text/plain"],
+    InputFormat.METS_GBS: ["application/mets+xml"],
     InputFormat.JSON_DOCLING: ["application/json"],
     InputFormat.AUDIO: ["audio/x-wav", "audio/mpeg", "audio/wav", "audio/mp3"],
 }

docling/datamodel/document.py

@@ -1,6 +1,7 @@
 import csv
 import logging
 import re
+import tarfile
 from collections.abc import Iterable
 from enum import Enum
 from io import BytesIO
@@ -314,6 +315,10 @@ class _DocumentConversionInput(BaseModel):
                 elif objname.endswith(".pptx"):
                     mime = "application/vnd.openxmlformats-officedocument.presentationml.presentation"
 
+        if mime is not None and mime.lower() == "application/gzip":
+            if detected_mime := _DocumentConversionInput._detect_mets_gbs(obj):
+                mime = detected_mime
+
         mime = mime or _DocumentConversionInput._detect_html_xhtml(content)
         mime = mime or _DocumentConversionInput._detect_csv(content)
         mime = mime or "text/plain"
@@ -457,3 +462,24 @@ class _DocumentConversionInput(BaseModel):
             return None
 
         return None
+
+    @staticmethod
+    def _detect_mets_gbs(
+        obj: Union[Path, DocumentStream],
+    ) -> Optional[Literal["application/mets+xml"]]:
+        content = obj if isinstance(obj, Path) else obj.stream
+        tar: tarfile.TarFile
+        member: tarfile.TarInfo
+        with tarfile.open(
+            name=content if isinstance(content, Path) else None,
+            fileobj=content if isinstance(content, BytesIO) else None,
+            mode="r:gz",
+        ) as tar:
+            for member in tar.getmembers():
+                if member.name.endswith(".xml"):
+                    file = tar.extractfile(member)
+                    if file is not None:
+                        content_str = file.read().decode(errors="ignore")
+                        if "http://www.loc.gov/METS/" in content_str:
+                            return "application/mets+xml"
+        return None

docling/datamodel/pipeline_options_vlm_model.py

@@ -1,5 +1,5 @@
 from enum import Enum
-from typing import Any, Callable, Dict, List, Literal, Optional, Union
+from typing import Any, Dict, List, Literal, Optional
 
 from docling_core.types.doc.page import SegmentedPage
 from pydantic import AnyUrl, BaseModel
@@ -10,11 +10,17 @@ from docling.datamodel.accelerator_options import AcceleratorDevice
 
 class BaseVlmOptions(BaseModel):
     kind: str
-    prompt: Union[str, Callable[[Optional[SegmentedPage]], str]]
+    prompt: str
     scale: float = 2.0
     max_size: Optional[int] = None
     temperature: float = 0.0
 
+    def build_prompt(self, page: Optional[SegmentedPage]) -> str:
+        return self.prompt
+
+    def decode_response(self, text: str) -> str:
+        return text
+
 
 class ResponseFormat(str, Enum):
     DOCTAGS = "doctags"

docling/document_converter.py

@@ -5,7 +5,9 @@ import threading
 import time
 from collections.abc import Iterable, Iterator
 from concurrent.futures import ThreadPoolExecutor
+from datetime import datetime
 from functools import partial
+from io import BytesIO
 from pathlib import Path
 from typing import Dict, List, Optional, Tuple, Type, Union
 
@@ -18,6 +20,7 @@ from docling.backend.docling_parse_v4_backend import DoclingParseV4DocumentBacke
 from docling.backend.html_backend import HTMLDocumentBackend
 from docling.backend.json.docling_json_backend import DoclingJSONBackend
 from docling.backend.md_backend import MarkdownDocumentBackend
+from docling.backend.mets_gbs_backend import MetsGbsDocumentBackend
 from docling.backend.msexcel_backend import MsExcelDocumentBackend
 from docling.backend.mspowerpoint_backend import MsPowerpointDocumentBackend
 from docling.backend.msword_backend import MsWordDocumentBackend
@@ -157,6 +160,9 @@ def _get_default_option(format: InputFormat) -> FormatOption:
         InputFormat.XML_JATS: FormatOption(
             pipeline_cls=SimplePipeline, backend=JatsDocumentBackend
         ),
+        InputFormat.METS_GBS: FormatOption(
+            pipeline_cls=StandardPdfPipeline, backend=MetsGbsDocumentBackend
+        ),
         InputFormat.IMAGE: FormatOption(
             pipeline_cls=StandardPdfPipeline, backend=DoclingParseV4DocumentBackend
         ),
@@ -275,6 +281,34 @@ class DocumentConverter:
                 "Conversion failed because the provided file has no recognizable format or it wasn't in the list of allowed formats."
             )
 
+    @validate_call(config=ConfigDict(strict=True))
+    def convert_string(
+        self,
+        content: str,
+        format: InputFormat,
+        name: Optional[str],
+    ) -> ConversionResult:
+        name = name or datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+
+        if format == InputFormat.MD:
+            if not name.endswith(".md"):
+                name += ".md"
+
+            buff = BytesIO(content.encode("utf-8"))
+            doc_stream = DocumentStream(name=name, stream=buff)
+
+            return self.convert(doc_stream)
+        elif format == InputFormat.HTML:
+            if not name.endswith(".html"):
+                name += ".html"
+
+            buff = BytesIO(content.encode("utf-8"))
+            doc_stream = DocumentStream(name=name, stream=buff)
+
+            return self.convert(doc_stream)
+        else:
+            raise ValueError(f"format {format} is not supported in `convert_string`")
+
     def _convert(
         self, conv_input: _DocumentConversionInput, raises_on_error: bool
     ) -> Iterator[ConversionResult]:

docling/models/api_vlm_model.py

@@ -53,11 +53,7 @@ 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
-
+                    prompt = self.vlm_options.build_prompt(page.parsed_page)
                     page_tags = api_image_request(
                         image=hi_res_image,
                         prompt=prompt,
@@ -67,6 +63,7 @@ class ApiVlmModel(BasePageModel):
                         **self.params,
                     )
 
+                    page_tags = self.vlm_options.decode_response(page_tags)
                     page.predictions.vlm_response = VlmPrediction(text=page_tags)
 
                 return page

docling/models/vlm_models_inline/hf_transformers_model.py

@@ -135,10 +135,7 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
                     )
 
                     # Define prompt structure
-                    if callable(self.vlm_options.prompt):
-                        user_prompt = self.vlm_options.prompt(page.parsed_page)
-                    else:
-                        user_prompt = self.vlm_options.prompt
+                    user_prompt = self.vlm_options.build_prompt(page.parsed_page)
                     prompt = self.formulate_prompt(user_prompt)
 
                     inputs = self.processor(
@@ -166,6 +163,7 @@ class HuggingFaceTransformersVlmModel(BasePageModel, HuggingFaceModelDownloadMix
                     _log.debug(
                         f"Generated {num_tokens} tokens in time {generation_time:.2f} seconds."
                     )
+                    generated_texts = self.vlm_options.decode_response(generated_texts)
                     page.predictions.vlm_response = VlmPrediction(
                         text=generated_texts,
                         generation_time=generation_time,

docling/models/vlm_models_inline/mlx_model.py

@@ -35,9 +35,9 @@ class HuggingFaceMlxModel(BasePageModel, HuggingFaceModelDownloadMixin):
 
         if self.enabled:
             try:
-                from mlx_vlm import generate, load  # type: ignore
+                from mlx_vlm import generate, load, stream_generate  # type: ignore
                 from mlx_vlm.prompt_utils import apply_chat_template  # type: ignore
-                from mlx_vlm.utils import load_config, stream_generate  # type: ignore
+                from mlx_vlm.utils import load_config  # type: ignore
             except ImportError:
                 raise ImportError(
                     "mlx-vlm is not installed. Please install it via `pip install mlx-vlm` to use MLX VLM models."
@@ -84,10 +84,7 @@ 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
+                    user_prompt = self.vlm_options.build_prompt(page.parsed_page)
                     prompt = self.apply_chat_template(
                         self.processor, self.config, user_prompt, num_images=1
                     )
@@ -142,6 +139,7 @@ class HuggingFaceMlxModel(BasePageModel, HuggingFaceModelDownloadMixin):
                     _log.debug(
                         f"{generation_time:.2f} seconds for {len(tokens)} tokens ({len(tokens) / generation_time} tokens/sec)."
                     )
+                    page_tags = self.vlm_options.decode_response(page_tags)
                     page.predictions.vlm_response = VlmPrediction(
                         text=page_tags,
                         generation_time=generation_time,