feautre(extract:docling): support docling perf settings

extract-core/extract_core/docling_.py

@@ -20,8 +20,14 @@
     TableStructureOptions,
     ThreadedPdfPipelineOptions,
 )
+from docling.datamodel.settings import (
+    BatchConcurrencySettings,
+    DebugSettings,
+    InferenceSettings,
+)
 from icij_common.pydantic_utils import (
     merge_configs,
+    safe_copy,
     tagged_union,
     to_lower_snake_case,
 )
@@ -258,12 +264,19 @@ def _default_format_opts() -> dict[InputFormat, DoclingFormatOption]:
     }
 
 
+class DoclingSettings(BaseModel):
+    perf: BatchConcurrencySettings = Field(default_factory=BatchConcurrencySettings)
+    debug: DebugSettings = Field(default_factory=DebugSettings)
+    inference: InferenceSettings = Field(default_factory=InferenceSettings)
+
+
 class DoclingPipelineConfig(BasePipelineConfig):
     pipeline: ClassVar[PipelineType] = Field(frozen=True, default=PipelineType.DOCLING)
 
     format_options: dict[InputFormat, DoclingFormatOption] = Field(
         default_factory=_default_format_opts
     )
+    settings: DoclingSettings = Field(default_factory=DoclingSettings)
 
     @classmethod
     @cache
@@ -276,3 +289,7 @@ def supported_exts(cls) -> set[SupportedExt]:
             for ext in FormatToExtensions[f]:
                 supported.add(SupportedExt(f".{ext.lower()}"))
         return supported
+
+    def with_setting(self, settings: DoclingSettings) -> "DoclingPipelineConfig":
+        update = {"settings": settings}
+        return safe_copy(self, update=update)

extract-python/README.md

@@ -0,0 +1 @@
+ô

extract-python/extract_python/docling_.py

@@ -10,6 +10,7 @@
 
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import PipelineOptions
+from docling.datamodel.settings import scoped
 from docling.document_converter import DocumentConverter, FormatOption
 
 # TODO: this is long to load improve it
@@ -62,16 +63,23 @@ def __init__(self, config: DoclingPipelineConfig):
     async def extract_content(
         self, docs: Iterable[InputDoc], output_format: OutputFormat, output_path: Path
     ) -> AsyncGenerator[Result, None]:
-        docs, path_or_streams = map_and_preserve(_to_docling, docs)
-        outputs = self._converter.convert_all(path_or_streams, raises_on_error=False)
-
-        sentinel = object()
-        while True:
-            res = await asyncio.to_thread(next, outputs, sentinel)
-            if res is sentinel:
-                return
-            doc = next(docs)
-            yield _to_result(res, doc, output_format, output_path=output_path)
+        settings = self._config.settings
+        logger.info("starting extraction with settings: %s", settings)
+        with scoped(
+            perf=settings.perf, debug=settings.debug, inference=settings.inference
+        ):
+            docs, path_or_streams = map_and_preserve(_to_docling, docs)
+            outputs = self._converter.convert_all(
+                path_or_streams, raises_on_error=False
+            )
+
+            sentinel = object()
+            while True:
+                res = await asyncio.to_thread(next, outputs, sentinel)
+                if res is sentinel:
+                    return
+                doc = next(docs)
+                yield _to_result(res, doc, output_format, output_path=output_path)
 
 
 def _to_docling(docs: Iterable[InputDoc]) -> Iterator["Path | DocumentStream"]: