version 3

app.py

@@ -3,7 +3,6 @@ from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipe
 from peft import PeftModel
 import torch
 
-# Load model and tokenizer with adapter
 @st.cache_resource
 def load_model():
     base_model = "Qwen/Qwen3-0.6B"
@@ -20,11 +19,9 @@ def load_model():
     model = PeftModel.from_pretrained(base, adapter_path)
     model = model.merge_and_unload()
 
-    # ✅ Text classification pipeline
     pipe = pipeline("text-classification", model=model, tokenizer=tokenizer)
     return pipe
 
-# Load pipeline once
 classifier = load_model()
 
 # Streamlit UI
@@ -33,19 +30,21 @@ text = st.text_area("Enter a news statement or claim:", height=200)
 
 if st.button("Check"):
     with st.spinner("Analyzing..."):
-        # Get classification result
         result = classifier(text)[0]
         label = result['label']
         score = result['score']
 
-        # Optional: format label nicely
-        if "1" in label or "POSITIVE" in label.upper():
+        st.write("🔎 Raw label:", label)  # Debug print
+
+        if label == "LABEL_1":
             verdict = "Real"
             emoji = "✅"
-        else:
+        elif label == "LABEL_0":
             verdict = "Fake"
             emoji = "❌"
+        else:
+            verdict = f"Unclear ({label})"
+            emoji = "🤔"
 
-        # Show result
         st.subheader("Prediction")
         st.success(f"{emoji} The statement is likely: **{verdict}** (confidence: {score:.2f})")