当前位置：首页 > news >正文

记录paddlepaddle-gpu安装

news 来源：原创 2025/5/20 23:56:28

背景

由于最近需要使用paddleocr，因此需要安装依赖paddlepaddle-gpu，不管怎么安装cuda11.6-11.8安装了一遍，都无法正常安装成功。如下所示：

环境：wsl2+linux18.04

>>> import paddle
>>> paddle.utils.run_check()

报错如下：

  PreconditionNotMetError: Cannot load cudnn shared library. Cannot invoke method cudnnGetVersion.
      [Hint: cudnn_dso_handle should not be null.] (at ../paddle/phi/backends/dynload/cudnn.cc:64)
      [operator < fill_constant > error]

尝试

https://blog.csdn.net/qq451882471/article/details/106967942
https://blog.csdn.net/weixin_44065323/article/details/110082334

根据上面两位大佬的做法，无法解决；
怀疑是cuda和cudnn版本不一致，各种版本都试了一轮，也都无法解决；

docker

对paddle也是无语了，兼容性做的如此之差，谁还想继续使用下去（比torch、transformers库安装体验差多了），最后使用docker进行安装
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/docker/linux-docker.html

docker pull

nvidia-docker pull registry.baidubce.com/paddlepaddle/paddle:2.5.2-gpu-cuda11.7-cudnn8.4-trt8.4

docker run

 docker rundocker run --gpus=all --name docker_paddle -it -v /mnt:/mnt registry.baidub
ce.com/paddlepaddle/paddle:2.5.2-gpu-cuda11.7-cudnn8.4-trt8.4 /bin/bash --name paddle_docker -it -v /mnt:/mnt registry.baidubce.com/paddlepaddle/paddle:2.5.2-gpu-cuda11.7-cudnn8.4-trt8.4 /bin/bash

测试成功

λ 9d35c036063b /home python
Python 3.7.14 (default, Sep  8 2022, 00:06:44)
[GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import paddle
grep: warning: GREP_OPTIONS is deprecated; please use an alias or script
>>> paddle.utils.run_check()
Running verify PaddlePaddle program ...
I1104 10:41:01.608309    12 interpretercore.cc:237] New Executor is Running.
W1104 10:41:01.608773    12 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 8.9, Driver API Version: 12.3, Runtime API Version: 11.7
W1104 10:41:02.405509    12 gpu_resources.cc:149] device: 0, cuDNN Version: 8.4.
I1104 10:41:12.111374    12 interpreter_util.cc:518] Standalone Executor is Used.
PaddlePaddle works well on 1 GPU.
PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
>>>