Update app.py

app.py

@@ -95,12 +95,15 @@ def respond(
     print(messages)
     response = ""
         
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
+    for message in client.chat.completions.create( #client.chat_completion(
+        messages=messages,
+        streamer=streamer,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
         stream=True,
         temperature=temperature,
         top_p=top_p,
+        eos_token_id=terminators,
     ):
         token = message.choices[0].delta.content