参考链接:
import paddle # required: gpu paddle.set_device("gpu") #tensor = paddle.randn([512, 512, 512], "float") #del tensor paddle.device.cuda.empty_cache()