Update app.py

app.py

@@ -21,10 +21,10 @@ def generate_text(input_text, history):
         input_text_with_history += f"Q: {input_text} \n A:"
     print("new input", input_text_with_history)
     output = llm(input_text_with_history, max_tokens=1024, stop=["Q:", "\n"], stream=True)
-    history =["init",input_text_with_history]
     for out in output:
      stream = copy.deepcopy(out)
      yield print(stream["choices"][0]["text"])   
+    history =["init",input_text_with_history]
     
 
 demo = gr.ChatInterface(generate_text)