Update app.py

app.py

@@ -98,6 +98,9 @@ def respond(
             **parameters  # Pass all other parameters
         )
         
+        # Print a starting message for token streaming
+        print("Received tokens: ", end="", flush=True)
+        
         # Process the streaming response
         for chunk in stream:
             if hasattr(chunk, 'choices') and len(chunk.choices) > 0:
@@ -105,9 +108,13 @@ def respond(
                 if hasattr(chunk.choices[0], 'delta') and hasattr(chunk.choices[0].delta, 'content'):
                     token_text = chunk.choices[0].delta.content
                     if token_text:
-                        print(f"Received token: {token_text}")
+                        # Print tokens inline without newlines
+                        print(token_text, end="", flush=True)
                         response += token_text
                         yield response
+        
+        # Print a newline at the end of all tokens
+        print()
     except Exception as e:
         print(f"Error during inference: {e}")
         response += f"\nError: {str(e)}"