Update app.py

app.py

@@ -46,7 +46,7 @@ model = model.eval()
 def stream_chat(message: str, history: list, temperature: float, max_new_tokens: int, top_p: float, top_k: int, penalty: float):
     print(history)
     
-    resp = model.stream_chat(
+    for resp, history in model.stream_chat(
         tokenizer,
         query = message,
         history = history,
@@ -55,9 +55,9 @@ def stream_chat(message: str, history: list, temperature: float, max_new_tokens:
         top_p = top_p,
         top_k = top_k,
         temperature = temperature,
-    )
-    print(resp)
-    yield resp
+    ):
+        print(resp)
+        yield resp
 
 
 chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)