Update app.py

app.py

@@ -1,7 +1,6 @@
 from huggingface_hub import InferenceClient
 import gradio as gr
 import random
-client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
      
 from prompts import GAME_MASTER, COMPRESS_HISTORY, ADJUST_STATS
 def format_prompt(message, history):
@@ -22,6 +21,8 @@ top_p=0.95
 repetition_penalty=2.0
 
 def compress_history(history,temperature=temperature,top_p=top_p,repetition_penalty=repetition_penalty):
+    client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+    
     print("COMPRESSING")
     formatted_prompt=f"{COMPRESS_HISTORY.format(history=history)}"
     generate_kwargs = dict(
@@ -42,6 +43,8 @@ def compress_history(history,temperature=temperature,top_p=top_p,repetition_pena
 MAX_HISTORY=100
 
 def generate(prompt, history,max_new_tokens,health,temperature=temperature,top_p=top_p,repetition_penalty=repetition_penalty):
+    client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+    
     temperature = float(temperature)
     if temperature < 1e-2:
         temperature = 1e-2