Update app.py

app.py

@@ -26,12 +26,12 @@ VERBOSE = True
 MAX_HISTORY = 125
 
 def format_prompt(message, history):
-    prompt = "<s>"
+    messages = []
     for user_prompt, bot_response in history:
-        prompt += f"[INST] {user_prompt} [/INST]"
-        prompt += f" {bot_response}</s> "
-    prompt += f"[INST] {message} [/INST]"
-    return prompt
+        messages.append({"role": "user", "content": user_prompt})
+        messages.append({"role": "assistant", "content": bot_response})
+    messages.append({"role": "user", "content": message})
+    return messages
 
 # Define current date/time
 date_time_str = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
@@ -136,12 +136,12 @@ def respond(
     global client
 
     # Format the prompt
-    prompt = format_prompt(message, history)
+    messages = format_prompt(message, history)
 
     # Generate response using the InferenceClient
     response = ""
     for message in client.chat_completion(
-        prompt,
+        messages=messages,
         max_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
@@ -182,4 +182,4 @@ demo = gr.ChatInterface(
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)