版本paddlepaddle-gpu 2.3.1.post116，利用pycharm远程执行create_predictor函数，报错exit code 139

bug描述 Describe the Bug

问题描述

在Ubuntu20种安装了paddle运行环境，通过pycharm远程方式进行paddle库调用。
在调用mask_detector = hub.Module(name="pyramidbox_lite_mobile")加载模型时发现，直接在服务器上运行，可以执行成功，但是通过pycharm方式远程执行时，执行失败，报错，Process finished with exit code 139。
后经定位发现，pycharm远程方式只要执行到create_predictor方法，都会报同样的问题。直接在服务器上运行不会。

在另一个paddlepaddle-gpu 2.0.2.post100、paddlehub 2.0.2、cuda 10.0的环境下，也不存在该问题

运行环境

paddle 版本：paddlepaddle-gpu 2.3.1.post116
paddle hub版本：2.2.0
cuda 版本：cuda_11.6
操作系统：Ubuntu 20.04 LTS （wsl方式启动）

异常代码

    from paddle.inference import Config
    from paddle.inference import create_predictor
    default_pretrained_model_path = "/root/.paddlehub/modules/pyramidbox_lite_server/pyramidbox_lite_server_face_detection/"
    cpu_config = Config(default_pretrained_model_path)
    cpu_config.disable_glog_info()
    cpu_config.disable_gpu()
    cpu_predictor = create_predictor(cpu_config)

其他补充信息 Additional Supplementary Information

No response

Start with the Python entry points Config and create_predictor shown in the report, then reproduce the call under Ubuntu 20.04 in WSL through PyCharm and directly on the server. Compare the paddlepaddle-gpu 2.3.1.post116 environment with the reported 2.0.2.post100 setup and determine what causes exit code 139; done means the remote invocation no longer crashes or the incompatibility is documented.