Thanks what size gpu would you need to fine tune or do an inference?