Does running this model require a lot of GPU memory? Will using an A100 GPU still result in an out-of-memory error?