Update ai_text_detector_valid_final.py

ai_text_detector_valid_final.py

@@ -48,41 +48,33 @@ def clean_text(text: str) -> str:
     return text.strip()
 
 def classify_szegedai(text: str):
-    """ModernBERT ensemble detector with human label boost"""
     cleaned_text = clean_text(text)
     if not cleaned_text.strip():
         return {"error": "Empty text"}
 
     inputs = tokenizer_modernbert(cleaned_text, return_tensors="pt", truncation=True, padding=True).to(device)
-    
     with torch.no_grad():
         logits_1 = model_1(**inputs).logits
         logits_2 = model_2(**inputs).logits
         logits_3 = model_3(**inputs).logits
 
-        # Softmax probabilities
         probs1 = torch.softmax(logits_1, dim=1)
         probs2 = torch.softmax(logits_2, dim=1)
         probs3 = torch.softmax(logits_3, dim=1)
 
-        # Boost the human label slightly to reduce false positives
         human_index = 24
         for p in [probs1, probs2, probs3]:
-            p[:, human_index] *= 1.2
-            p = p / p.sum(dim=1, keepdim=True)  # re-normalize
+            p[:, human_index] *= 2.0  # Boost human label
+            p = p / p.sum(dim=1, keepdim=True)
 
-        # Ensemble average
         probs = (probs1 + probs2 + probs3) / 3
 
-    probs = probs[0]
-    ai_probs = probs.clone()
-    ai_probs[human_index] = 0
-    ai_total_prob = ai_probs.sum().item() * 100
-    human_prob = 100 - ai_total_prob
+    human_prob = probs[0][human_index].item() * 100
+    ai_prob = 100 - human_prob
 
     return {
         "Human Probability": round(human_prob, 2),
-        "AI Probability": round(ai_total_prob, 2),
+        "AI Probability": round(ai_prob, 2),
     }
 
 # ---------------------------