Update app.py

app.py

@@ -34,13 +34,13 @@ def generate_and_parse(messages: list, temperature: float = 0.6,
     and parses it into thinking and answer parts.
     Decorated with @spaces.GPU for Zero GPU allocation.
     """
-    # Apply chat template WITHOUT enable_thinking to preserve thinking tags in history
-    prompt_text = tokenizer.apply_chat_template(
-        messages,
-        tokenize=False,
-        add_generation_prompt=True,
-        enable_thinking=False  # Changed to False to preserve <think> tags in context
-    )
+    # Build prompt manually to preserve <think> tags in context
+    prompt_text = ""
+    for msg in messages:
+        role = msg["role"]
+        content = msg["content"]
+        prompt_text += f"<|im_start|>{role}\n{content}<|im_end|>\n"
+    prompt_text += "<|im_start|>assistant\n"
 
     # --- CONSOLE DEBUG OUTPUT ---
     print("\n" + "="*50)