Update app.py

app.py

@@ -7,7 +7,7 @@ client = InferenceClient("Futuresony/future_ai_12_10_2024.gguf")
 def format_alpaca_prompt(history, user_input, system_prompt):
     """Formats input in Alpaca/LLaMA style with conversation history"""
     formatted_history = "\n".join([f"User: {h[0]}\nAssistant: {h[1]}" for h in history])
-    prompt = f"""{system_prompt}\n{formatted_history}\nUser: {user_input}\n"""
+    prompt = f"""{system_prompt}\n{formatted_history}\nUser: {user_input}\nAssistant:"""
     return prompt
 
 def respond(message, history, system_message, max_tokens, temperature, top_p):
@@ -21,7 +21,7 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     )
     
     # Extract only the response
-    cleaned_response = response.strip()
+    cleaned_response = response.strip().split("Assistant:")[-1].strip()
     
     # Update history
     history.append((message, cleaned_response))