Demo Fixes 6

Backend/server.py

@@ -112,6 +112,15 @@ def load_models():
             torch_dtype=torch.bfloat16
         )
         models.tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-1B")
+        
+        # Configure all special tokens
+        models.tokenizer.pad_token = models.tokenizer.eos_token
+        models.tokenizer.padding_side = "left"  # For causal language modeling
+
+        # Inform the model about the pad token
+        if hasattr(models.llm.config, "pad_token_id") and models.llm.config.pad_token_id is None:
+            models.llm.config.pad_token_id = models.tokenizer.pad_token_id
+        
         logger.info("Llama 3.2 model loaded successfully")
         socketio.emit('model_status', {'model': 'llm', 'status': 'loaded'})
         progress = 100
@@ -392,6 +401,11 @@ def process_audio_and_respond(session_id, data):
             prompt = f"{conversation_history}Assistant: "
             
             # Generate response with Llama
+            try:
+                # Ensure pad token is set
+                if models.tokenizer.pad_token is None:
+                    models.tokenizer.pad_token = models.tokenizer.eos_token
+                    
                 input_tokens = models.tokenizer(
                     prompt, 
                     return_tensors="pt",
@@ -417,6 +431,11 @@ def process_audio_and_respond(session_id, data):
                     generated_ids[0][input_ids.shape[1]:], 
                     skip_special_tokens=True
                 ).strip()
+            except Exception as e:
+                logger.error(f"Error generating response: {str(e)}")
+                import traceback
+                logger.error(traceback.format_exc())
+                response_text = "I'm sorry, I encountered an error while processing your request."
             
             # Synthesize speech
             with app.app_context():