Correction for runtime error

app.py

@@ -17,7 +17,6 @@ LABEL_COLS = ['anger', 'fear', 'joy', 'sadness', 'surprise']
 DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
 # --- CORRECTED FILE PATHS ---
-# These paths now correctly point to the 'model/' directory.
 MODEL_FILE = "Deploy/models/best_deberta_0.8558.pt" 
 THRESHOLD_FILE = "Deploy/models/optimal_thresholds_deberta.npy"
 
@@ -50,21 +49,22 @@ def load_global_assets():
 
     print(f"Loading assets from {MODEL_FILE}...")
     try:
-        # 🔑 FIX: Add use_fast=False to resolve the 'NoneType' object has no attribute 'endswith' error
-        # This forces the use of the "slow" tokenizer which correctly resolves the spm.model path.
+        # Load slow tokenizer explicitly to avoid 'NoneType' endswith error
+        # Alternative: from transformers import DebertaV2Tokenizer
         tokenizer = AutoTokenizer.from_pretrained(
             CONFIG['MODEL_NAME'], 
-            use_fast=False  # <--- CRITICAL FIX FOR DEBERTA-V3
-        ) 
-        
+            use_fast=False,
+            force_download=True,    # force fresh download to avoid cache issues
+            local_files_only=False
+        )
+
         model = DebertaClassifier(
             n_classes=CONFIG['OUTPUT_DIM'],
             model_name=CONFIG['MODEL_NAME'],
             dropout=CONFIG['DROPOUT']
         )
-        
+
         # Load the trained PyTorch state dict
-        # Ensure the file path is correct relative to the Space root directory
         state_dict = torch.load(MODEL_FILE, map_location=torch.device('cpu'))
         model.load_state_dict(state_dict)
         model.to(DEVICE)
@@ -77,26 +77,20 @@ def load_global_assets():
         print("Model and thresholds loaded successfully.")
     except Exception as e:
         print(f"Error loading model assets: {e}")
-        # Use gr.Warning instead of gr.Error to allow the app to initialize but show a warning
         raise gr.Error(f"Deployment failed to load assets. Check file paths. Error: {e}")
+
 # Call the loading function once before defining the interface
 load_global_assets()
 
-
 # --- 3. Gradio Interface Function ---
 def classify_emotion(text: str) -> str:
-    """
-    Takes user text, runs inference, applies thresholds, and returns
-    a formatted string of predicted emotions.
-    """
+    """Tokenizes input, runs model inference, applies thresholds, formats output."""
     if not text.strip():
         return "Please enter a sentence to classify."
     
-    # Check if the model failed to load during startup
     if model is None:
         return "Model failed to load during startup. Check the Space logs for errors."
 
-    # Tokenize the input text
     inputs = tokenizer(
         text, 
         return_tensors="pt", 
@@ -104,46 +98,35 @@ def classify_emotion(text: str) -> str:
         padding='max_length',
         max_length=CONFIG['MAX_LEN']
     )
-    
     input_ids = inputs['input_ids'].to(DEVICE)
     attention_mask = inputs['attention_mask'].to(DEVICE)
 
-    # Get model outputs
     with torch.no_grad():
         logits = model(input_ids=input_ids, attention_mask=attention_mask)
         
-    # Apply Sigmoid activation
     probabilities = torch.sigmoid(logits).squeeze(0).cpu().numpy()
-    
-    # Apply the custom label-specific thresholds
     predictions = probabilities > thresholds
     
-    # Map predictions back to your emotion labels
     predicted_emotions = [
         f"**{LABEL_COLS[i]}** ({probabilities[i]:.2f} > {thresholds[i]:.2f})" 
         for i, pred in enumerate(predictions) if pred
     ]
     
     if predicted_emotions:
-        # Format the output as a list for the user
         output_list = "\n".join([f"- {e}" for e in predicted_emotions])
         return f"**Detected Emotions:**\n{output_list}"
     else:
         return "No emotions were detected above the optimal thresholds."
 
-
 # --- 4. Define and Launch Gradio Interface ---
-# Input component: Textbox
 text_input = gr.Textbox(
     lines=5, 
     placeholder="Example: I was so furious when they broke my camera, but happy I had a backup.",
     label="Text Input"
 )
 
-# Output component: Textbox (set to display markdown for bold text and lists)
 text_output = gr.Markdown(label="Predicted Emotions")
 
-# Create the Gradio Interface
 gr.Interface(
     fn=classify_emotion,
     inputs=text_input,