Update mdr_pdf_parser.py

mdr_pdf_parser.py

@@ -1813,23 +1813,22 @@ class MDRLayoutReader:
 # In class MDRLayoutReader:
   def _get_model(self) -> LayoutLMv3ForTokenClassification | None:
     if self._model is None:
-      cache = mdr_ensure_directory(self._model_path) # This should be self._model_path / "layoutreader"
-      # Correct cache path for transformers
-      layoutreader_cache_dir = Path(self._model_dir) / "layoutreader" # Assuming _model_dir is the main one
-      mdr_ensure_directory(str(layoutreader_cache_dir))
+      # MODIFIED: Use self._model_path for the layoutreader's specific cache,
+      # and ensure it's a directory. self._model_path is passed during MDRLayoutReader init.
+      layoutreader_cache_dir = Path(self._model_path) # self._model_path is like "./mdr_models/layoutreader"
+      mdr_ensure_directory(str(layoutreader_cache_dir)) # Ensure this specific directory exists
 
       name = "microsoft/layoutlmv3-base"
       
       print(f"MDRLayoutReader: Attempting to load LayoutLMv3 model '{name}'. Cache dir: {layoutreader_cache_dir}")
       try:
         self._model = LayoutLMv3ForTokenClassification.from_pretrained(
-            name, # Use the HF model name
+            name, 
             cache_dir=str(layoutreader_cache_dir), 
-            local_files_only=False, # Allow download on first run
+            local_files_only=False, 
             num_labels=_MDR_MAX_LEN+1 
         )
-        # Explicitly move model to the determined device
-        self._model.to(torch.device(self._device)) # ENSURE THIS LINE IS PRESENT AND CORRECT
+        self._model.to(torch.device(self._device))
         self._model.eval()
         print(f"MDR LayoutReader model '{name}' loaded successfully on device: {self._model.device}.")
       except Exception as e: