Update app.py

app.py

@@ -82,36 +82,31 @@ def respond(message, history, level, max_tokens, temperature, top_p):
         return
     
     system_message = level_to_prompt(level)
-    messages = [{"role": "system", "content": system_message}]
-    
-    # Handle history based on its format
-    if history and isinstance(history[0], dict):
-        # New format (messages with role/content)
-        messages.extend(history)
-    else:
-        # Old format (tuples)
-        for user, bot in history:
-            if user:
-                messages.append({"role": "user", "content": user})
-            if bot:
-                messages.append({"role": "assistant", "content": bot})
-    
-    # Add current message
-    messages.append({"role": "user", "content": message})
     
     # Generate response
     response = ""
     try:
-        # Create a proper prompt format
-        prompt = ""
-        for msg in messages:
-            if msg["role"] == "system":
-                prompt += f"System: {msg['content']}\n"
-            elif msg["role"] == "user":
-                prompt += f"User: {msg['content']}\n"
-            elif msg["role"] == "assistant":
-                prompt += f"Assistant: {msg['content']}\n"
-        prompt += "Assistant: "
+        # Create a proper prompt format for instruction-following models
+        prompt = f"<|system|>\n{system_message}\n\n"
+        
+        # Add conversation history
+        if history:
+            for turn in history:
+                if isinstance(turn, dict):
+                    if turn.get("role") == "user":
+                        prompt += f"<|user|>\n{turn['content']}\n\n"
+                    elif turn.get("role") == "assistant":
+                        prompt += f"<|assistant|>\n{turn['content']}\n\n"
+                else:
+                    # Handle tuple format (user, assistant)
+                    user_msg, bot_msg = turn
+                    if user_msg:
+                        prompt += f"<|user|>\n{user_msg}\n\n"
+                    if bot_msg:
+                        prompt += f"<|assistant|>\n{bot_msg}\n\n"
+        
+        # Add current user message
+        prompt += f"<|user|>\n{message}\n\n<|assistant|>\n"
         
         # Generate response with streaming
         for token in client.text_generation(
@@ -121,11 +116,15 @@ def respond(message, history, level, max_tokens, temperature, top_p):
             temperature=temperature, 
             top_p=top_p,
             do_sample=True,
-            return_full_text=False
+            return_full_text=False,
+            stop_sequences=["<|user|>", "<|system|>"]  # Stop if model tries to continue conversation
         ):
             if token:  # Handle None tokens
-                response += token
-                yield response
+                # Clean up any unwanted tokens
+                token = token.replace("<|user|>", "").replace("<|system|>", "").replace("<|assistant|>", "")
+                if token.strip():  # Only add non-empty tokens
+                    response += token
+                    yield response
                 
     except Exception as e:
         error_msg = str(e)
@@ -151,6 +150,15 @@ with gr.Blocks(css=css, title="French Tutor") as demo:
         gr.Markdown("✅ **Status**: Connected to AI service")
     
     with gr.Column(elem_id="chat-panel"):
+        gr.Markdown("""
+        ### 💬 **Try these example conversations:**
+        - "Bonjour! Comment allez-vous?" (Hello! How are you?)
+        - "Can you help me conjugate the verb 'être'?"
+        - "What's the difference between 'tu' and 'vous'?"
+        - "Tell me about French culture"
+        - "How do I order food in a French restaurant?"
+        """)
+        
         with gr.Accordion("⚙️ Advanced Settings", open=False):
             level = gr.Dropdown(
                 choices=["A1", "A2", "B1", "B2", "C1", "C2"],