Update handler.py

handler.py

@@ -32,7 +32,16 @@ class EndpointHandler:
         try:
             # Load tokenizer and model explicitly to ensure correct loading
             tokenizer = AutoTokenizer.from_pretrained(model_path)
-            model = AutoModelForQuestionAnswering.from_pretrained(model_path)
+            model, loading_info = AutoModelForQuestionAnswering.from_pretrained(
+                model_path,
+                output_loading_info=True
+            )
+
+            logger.warning("RECCON load info - missing_keys: %s", loading_info.get("missing_keys"))
+            logger.warning("RECCON load info - unexpected_keys: %s", loading_info.get("unexpected_keys"))
+            logger.warning("RECCON load info - error_msgs: %s", loading_info.get("error_msgs"))
+            logger.warning("Loaded model class: %s", model.__class__.__name__)
+            logger.warning("Loaded model name_or_path: %s", getattr(model.config, "_name_or_path", None))
 
             # Initialize the pipeline
             # top_k=20 matches your previous 'n_best_size=20' logic