changed to stream

app.py

@@ -14,14 +14,18 @@ def respond(message, history):
         
     messages.append({"role" : "user", "content" : message})
 
-    response = client.chat_completion(messages, max_tokens = 50, temperature= .1, top_p= 0.7)
-    # max tokens = change the length of the response 
-    # temp = between 0-2
-    # top-p =  between 0-1
-
-    print(response["choices"][0]["message"]["content"].strip())
-    
-    return response["choices"][0]["message"]["content"].strip()
+    response = ""
+    for message in client.chat_completion(
+        messages, max_tokens = 100, stream=True
+        #temperature= .1, top_p= 0.7)
+        # max tokens = change the length of the response 
+        # temp = between 0-2
+        # top-p =  between 0-1
+    ):
+        token = messages.choices[0].delta.cotent
+        response =+ token
+        
+        yield response
 
 def random_message(message, history):
     choices = ["it is likely", "absolutely not", "try again", "without a doubt", "outlook good", "signs point to it", "very unlikely", "quite doubtful"]