单机多卡环境中，x.clone() 报错 NotImplementedError

[LemonNoel]

bug描述 Describe the Bug

复现环境：

• paddlepaddle-gpu=develop (CUDA=11.2, CUDNN=8.1)
• paddlenlp=develop
• 代码路径：https://github.com/PaddlePaddle/PaddleNLP/tree/develop/model_zoo/ernie-health
• 执行脚本：

  unset CUDA_VISIBLE_DEVICES
  python -m paddle.distributed.launch --gpus "0,1" run_pretrain.py \
  --input_dir ./data \
  --output_dir ./output \
  --learning_rate 1e-7 \
  --batch_size 10 \
  --adam_epsilon 1e-8 \
  --weight_decay 1e-2 \
  --warmup_steps 10000 \
  --max_steps 1000000 \
  --save_steps 10000 \
  --logging_steps 1 \
  --seed 1000 \
  --use_amp

• 报错信息 log/workerlog.1

  25939 ------------------------------------------------
  25940 I1221 11:53:31.671428 19119 tcp_utils.cc:130] Successfully connected to 10.21.226.184:44296
  25941 W1221 11:53:34.340793 19119 gpu_resources.cc:61] Please NOTE: device: 1, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 11.2
  25942 W1221 11:53:34.344606 19119 gpu_resources.cc:91] device: 1, cuDNN Version: 8.1.
  25943 ^[[32m[2022-12-21 11:53:34,583] [    INFO]^[[0m - Already cached /ssd2/.paddlenlp/models/ernie-health-chinese/vocab.txt^[[0m
  25944 ^[[32m[2022-12-21 11:53:34,598] [    INFO]^[[0m - tokenizer config file saved in /ssd2/.paddlenlp/models/ernie-health-chinese/tokenizer_config.json^[[0m
  25945 ^[[32m[2022-12-21 11:53:34,598] [    INFO]^[[0m - Special tokens file saved in /ssd2/.paddlenlp/models/ernie-health-chinese/special_tokens_map.json^[[0m
  25946 ^[[32m[2022-12-21 11:53:35,996] [    INFO]^[[0m - start load data : 2022-12-21 11:53:35^[[0m
  25947 ^[[32m[2022-12-21 11:53:36,017] [    INFO]^[[0m - load data done, total : 0.021075010299682617 s^[[0m
  25948 ^[[32m[2022-12-21 11:53:36,031] [    INFO]^[[0m - start train : 2022-12-21 11:53:36^[[0m
  25949 I1221 11:53:36.031898 19119 reader_py.cc:406] init_lod_tensor_blocking_queue
  25950 I1221 11:53:36.032111 19119 buffered_reader.cc:50] BufferedReader
  25951 Exception in thread Thread-1:
  25952 Traceback (most recent call last):
  25953   File "/ssd2/anaconda3/envs/black/lib/python3.8/threading.py", line 932, in _bootstrap_inner
  25954     self.run()
  25955   File "/ssd2/anaconda3/envs/black/lib/python3.8/threading.py", line 870, in run
  25956     self._target(*self._args, **self._kwargs)
  25957   File "/ssd2/anaconda3/envs/black/lib/python3.8/site-packages/paddle/fluid/dataloader/dataloader_iter.py", line 247, in _thread_loop
  25958     batch = self._dataset_fetcher.fetch(
  25959   File "/ssd2/anaconda3/envs/black/lib/python3.8/site-packages/paddle/fluid/dataloader/fetcher.py", line 139, in fetch
  25960     data = self.collate_fn(data)
  25961   File "/ssd2/prompt/PaddleNLP/model_zoo/ernie-health/dataset.py", line 102, in __call__
  25962     masked_input_ids_a, input_ids_a, labels_a = self.mask_tokens(data)
  25963   File "/ssd2/prompt/PaddleNLP/model_zoo/ernie-health/dataset.py", line 123, in mask_tokens
  25964     masked_token_ids = paddle.clone(token_ids)
  25965   File "/ssd2/anaconda3/envs/black/lib/python3.8/site-packages/paddle/tensor/creation.py", line 2123, in clone
  25966     return x.clone()
  25967 NotImplementedError: (Unimplemented) Place Place(gpu:0) is not supported. Please check that your paddle compiles with WITH_GPU, WITH_XPU, WITH_IPU, WITH_MLU or WITH_ASCEND_      CL option or check that your train process set the correct device id if you use Executor. (at /paddle/paddle/fluid/platform/device_context.cc:122)

其他补充信息 Additional Supplementary Information

No response

[paddle-bot[bot]]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[zhaoyinglia]

你好。请问编译 paddle 的时候是否开启了 DWITH_GPU 选项

[LemonNoel]

你好，是直接从官网拉取的whl包