Update run.py

run.py

@@ -215,6 +215,7 @@ def response(message, history,customSysPrompt, genre, hfToken):
   if((onPrem==False) & (hfToken.startswith("hf_"))): # use HF-hub with custom token if token is provided
     from huggingface_hub import InferenceClient
     import gradio as gr
+    global client
     client = InferenceClient(
       model=myModel,
       token=hfToken