make cpu memory usage low

app.py

@@ -46,6 +46,7 @@ def model_call(question, document, answer):
         do_sample=True,
         attention_mask=attention_mask,
         pad_token_id=tokenizer.eos_token_id,
+        low_cpu_mem_usage=True
     )
     with torch.no_grad():
         outputs = model.generate(**generate_kwargs)