updatE

app.py

@@ -34,11 +34,27 @@ pipeline_model = pipeline(
 def generate_with_pipeline(messages, max_new_tokens=256, temperature=0.7, top_p=0.9, top_k=50, repetition_penalty=1.0):
     """Generate response using the pipeline with messages format"""
     # Apply chat template
-    prompt = pipeline_model.tokenizer.apply_chat_template(
-        messages,
-        tokenize=False,
-        add_generation_prompt=True
-    )
+    try:
+        prompt = pipeline_model.tokenizer.apply_chat_template(
+            messages,
+            tokenize=False,
+            add_generation_prompt=True
+        )
+    except Exception as template_error:
+        print(f"Template application error: {template_error}")
+        # Fallback: manually format messages
+        prompt = ""
+        for msg in messages:
+            if msg['role'] == 'system':
+                prompt += f"System: {msg['content']}\n\n"
+            elif msg['role'] == 'user':
+                prompt += f"User: {msg['content']}\n"
+            elif msg['role'] == 'assistant':
+                prompt += f"Assistant: {msg['content']}\n"
+        prompt += "Assistant: "
+
+    # Debug: print final prompt
+    print(f"Final prompt preview: {prompt[:200]}...")
 
     outputs = pipeline_model(
         prompt,
@@ -64,26 +80,34 @@ def generate_response(message, history, max_new_tokens, temperature, top_p, top_
         max_new_tokens, temperature, top_p, top_k, repetition_penalty: Generation parameters
     """
     try:
-        # Build messages list starting with system prompt
-        messages = [{"role": "system", "content": DEFAULT_SYSTEM_PROMPT}]
-        
-        # Add conversation history
-        # When type="messages", history is a list of message dicts with 'role' and 'content'
+        # Build messages list - Gemma template expects alternating user/assistant after system
+        messages = []
+
+        # Add system message first (will be handled specially by the template)
+        messages.append({"role": "system", "content": DEFAULT_SYSTEM_PROMPT})
+
+        # Add conversation history (ensure alternating user/assistant)
         if history:
             for msg in history:
                 if isinstance(msg, dict) and 'role' in msg and 'content' in msg:
-                    messages.append({"role": msg['role'], "content": msg['content']})
-        
+                    # Convert 'assistant' role to 'model' for Gemma template if needed
+                    role = msg['role']
+                    if role == 'assistant':
+                        role = 'assistant'  # Keep as assistant, template converts to 'model'
+                    messages.append({"role": role, "content": msg['content']})
+
         # Add current user message
         if isinstance(message, dict):
             current_message = message.get("text", "") or message.get("content", "")
         else:
             current_message = str(message)
-        
+
         messages.append({"role": "user", "content": current_message})
         
         # Debug: print messages structure
         print(f"Messages sent to model: {len(messages)} messages")
+        for i, msg in enumerate(messages):
+            print(f"  Message {i}: role={msg['role']}, content_preview={msg['content'][:50]}...")
         
         # Generate response
         response = generate_with_pipeline(