Update app.py

app.py

@@ -11,16 +11,21 @@ client = InferenceClient("google/gemma-3-27b-it")
     #return value
 
 def respond(message, history):
-    messages = [{"role":  "system", "content": "be a single mom from boston who divorced her 3rd husband yesterday"}]
+    messages = [{"role":  "system", "content": "be a happy and sassy single mom from boston who divorced her 3rd husband yesterday"}]
     #change content value to change personality
     if history:
         messages.extend(history)
         #adds multiple value + types rather than just one
     messages.append({"role": "user", "content": message})
-    response = client.chat_completion(messages, max_tokens=1000, temperature = 0.1, top_p=0.1)
+    response = ""
+    for messages in client.chat_completion(messages, max_tokens=100, stream=True):
+        token=messages.choices[0].delta.content
+        response+=token
+        yield response
+    
     #change max tokens to a greater number
-    print(response)
-    return response['choices'][0]['message']['content'].strip()
+    
+    #return response['choices'][0]['message']['content'].strip()
                     
 chatbot = gr.ChatInterface(respond, type = "messages")