Update app.py

app.py

@@ -199,19 +199,14 @@ def _load_model():
         num_labels = int(getattr(_model.config, "num_labels", 0) or 0)
         _NORM_LABELS_BY_IDX = [_normalize_label_text_only(id2label.get(i, f"LABEL_{i}")) for i in range(num_labels)]
         
-        print(f"Model loaded successfully. Labels: {_NORM_LABELS_BY_IDX}")
+        print(f"Model loaded successfully. Number of labels: {num_labels}")
+        print(f"Label mapping: {id2label}")
+        print(f"Normalized labels: {_NORM_LABELS_BY_IDX}")
 
 
 def _predict_texts(texts: List[str], include_preprocessing: bool = True) -> List[Dict]:
     """
-    Predict and return strictly model-authoritative outputs:
-    - label: normalized model label (PHISH/LEGIT or other model label uppercased)
-    - raw_label: original id2label string from model.config
-    - is_phish: boolean derived from normalized label
-    - score: probability of predicted class
-    - probs: dict of normalized label -> probability
-    - predicted_index: argmax index
-    - preprocessing: detailed preprocessing steps (if requested)
+    Predict and return strictly model-authoritative outputs with enhanced debugging.
     """
     _load_model()
     if not texts:
@@ -267,6 +262,8 @@ def _predict_texts(texts: List[str], include_preprocessing: bool = True) -> List
             "score": round(float(p[idx].item()), 4),
             "probs": {k: round(v, 4) for k, v in prob_map.items()},
             "predicted_index": idx,
+            "all_logits": [round(float(logits[i][j].item()), 4) for j in range(logits.shape[1])],  # DEBUG
+            "raw_probs": [round(float(p[j].item()), 4) for j in range(len(p))],  # DEBUG
         }
         
         # Add preprocessing info if requested