Update app.py

app.py

@@ -8,7 +8,9 @@ tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_PATH,
     trust_remote_code=True,
-    device_map="auto"
+    device_map="auto",
+    low_cpu_mem_usage=True,
+    quantization_config=None
 )
 
 pipe = pipeline(