Update app.py

app.py

@@ -14,7 +14,8 @@ dtype = torch.bfloat16
 
 quantization_config = BitsAndBytesConfig(load_in_8bit=True)
 
-def load_model():
+@spaces.GPU
+def moderate(user_input, assistant_response):
     tokenizer = AutoTokenizer.from_pretrained(model_id, token=huggingface_token)
     model = AutoModelForCausalLM.from_pretrained(
         model_id, 
@@ -24,12 +25,7 @@ def load_model():
         token=huggingface_token,
         low_cpu_mem_usage=True
     )
-    return tokenizer, model
-
-tokenizer, model = load_model()
 
-@spaces.GPU
-def moderate(user_input, assistant_response):
     chat = [
         {"role": "user", "content": user_input},
         {"role": "assistant", "content": assistant_response},