updated app again

app.py

@@ -12,15 +12,22 @@ llm = Llama(
     model_path=model_path,
     n_ctx=2048,
     n_gpu_layers=0,
+    chat_format=None,   # <-- CRITICAL: disable chat templates
 )
 
-def respond(prompt):
+def respond(user_question):
+    # Format prompt exactly like your training data
+    prompt = f"Q: {user_question}\nA:"
+    
     out = llm.create_completion(
         prompt=prompt,
         max_tokens=256,
         temperature=0.7,
         top_p=0.9,
     )
+    
+    # Return the generated answer
     return out["choices"][0]["text"]
 
 gr.Interface(fn=respond, inputs="text", outputs="text").launch()
+