Update paddle, and support invoice line item

packages/backend/backend/web/services/inference.py

@@ -42,6 +42,11 @@ class ServiceResult:
     visualization_path: Path | None = None
     errors: list[str] = field(default_factory=list)
 
+    # Business features (optional, populated when extract_line_items=True)
+    line_items: dict | None = None
+    vat_summary: dict | None = None
+    vat_validation: dict | None = None
+
 
 class InferenceService:
     """
@@ -74,6 +79,7 @@ class InferenceService:
         self._detector = None
         self._is_initialized = False
         self._current_model_path: Path | None = None
+        self._business_features_enabled = False
 
     def _resolve_model_path(self) -> Path:
         """Resolve the model path to use for inference.
@@ -95,12 +101,16 @@ class InferenceService:
 
         return self.model_config.model_path
 
-    def initialize(self) -> None:
-        """Initialize the inference pipeline (lazy loading)."""
+    def initialize(self, enable_business_features: bool = False) -> None:
+        """Initialize the inference pipeline (lazy loading).
+
+        Args:
+            enable_business_features: Whether to enable line items and VAT extraction
+        """
         if self._is_initialized:
             return
 
-        logger.info("Initializing inference service...")
+        logger.info(f"Initializing inference service (business_features={enable_business_features})...")
         start_time = time.time()
 
         try:
@@ -118,16 +128,18 @@ class InferenceService:
                 device="cuda" if self.model_config.use_gpu else "cpu",
             )
 
-            # Initialize full pipeline
+            # Initialize full pipeline with optional business features
             self._pipeline = InferencePipeline(
                 model_path=str(model_path),
                 confidence_threshold=self.model_config.confidence_threshold,
                 use_gpu=self.model_config.use_gpu,
                 dpi=self.model_config.dpi,
                 enable_fallback=True,
+                enable_business_features=enable_business_features,
             )
 
             self._is_initialized = True
+            self._business_features_enabled = enable_business_features
             elapsed = time.time() - start_time
             logger.info(f"Inference service initialized in {elapsed:.2f}s with model: {model_path}")
 
@@ -242,6 +254,7 @@ class InferenceService:
         pdf_path: Path,
         document_id: str | None = None,
         save_visualization: bool = True,
+        extract_line_items: bool = False,
     ) -> ServiceResult:
         """
         Process a PDF file and extract invoice fields.
@@ -250,12 +263,17 @@ class InferenceService:
             pdf_path: Path to PDF file
             document_id: Optional document ID
             save_visualization: Whether to save visualization
+            extract_line_items: Whether to extract line items and VAT info
 
         Returns:
             ServiceResult with extracted fields
         """
         if not self._is_initialized:
-            self.initialize()
+            self.initialize(enable_business_features=extract_line_items)
+        elif extract_line_items and not self._business_features_enabled:
+            # Reinitialize with business features if needed
+            self._is_initialized = False
+            self.initialize(enable_business_features=True)
 
         doc_id = document_id or str(uuid.uuid4())[:8]
         start_time = time.time()
@@ -263,8 +281,12 @@ class InferenceService:
         result = ServiceResult(document_id=doc_id)
 
         try:
-            # Run inference pipeline
-            pipeline_result = self._pipeline.process_pdf(pdf_path, document_id=doc_id)
+            # Run inference pipeline with optional business features
+            pipeline_result = self._pipeline.process_pdf(
+                pdf_path,
+                document_id=doc_id,
+                extract_line_items=extract_line_items,
+            )
 
             result.fields = pipeline_result.fields
             result.confidence = pipeline_result.confidence
@@ -288,6 +310,12 @@ class InferenceService:
                 for d in pipeline_result.raw_detections
             ]
 
+            # Include business features if extracted
+            if extract_line_items:
+                result.line_items = pipeline_result._line_items_to_json() if pipeline_result.line_items else None
+                result.vat_summary = pipeline_result._vat_summary_to_json() if pipeline_result.vat_summary else None
+                result.vat_validation = pipeline_result._vat_validation_to_json() if pipeline_result.vat_validation else None
+
             # Save visualization (render first page)
             if save_visualization and pipeline_result.raw_detections:
                 viz_path = self._save_pdf_visualization(pdf_path, doc_id)