limit context length to 1500

app.py

@@ -25,10 +25,12 @@ def ask(question, history, behavior):
             for i,content in enumerate(history + [question])
         ]
         length_messages = num_tokens_from_messages(messages)
+        """
         time_penalty = (length_messages-1000)//10
         if time_penalty>0:
             print(f"sleep for {time_penalty:.2f}s for too long a quest: {length_messages}")
             time.sleep(time_penalty)
+        """
         response = openai.ChatCompletion.create(
             model="gpt-3.5-turbo",
             messages=forget_long_term(messages)
@@ -62,7 +64,7 @@ def num_tokens_from_messages(messages, model="gpt-3.5-turbo"):
         raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.
 See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")
 
-def forget_long_term(messages, max_num_tokens=4000):
+def forget_long_term(messages, max_num_tokens=1500):
     while num_tokens_from_messages(messages)>max_num_tokens:
         if messages[0]["role"]=="system" and not len(messages[0]["content"]>=max_num_tokens):
             messages = messages[:1] + messages[2:]
@@ -194,5 +196,5 @@ with gr.Blocks() as demo:
     button_rtr.click(fn=retry, inputs=[txt, state, behavior], outputs=[txt, state, chatbot, downloadfile])
     button_clr.click(fn=lambda :([],[]), inputs=None, outputs=[chatbot, state])
 
-demo.queue(concurrency_count=3, max_size=10)
+#demo.queue(concurrency_count=3, max_size=10)
 demo.launch()