记录服务器上cuda和对应的paddlepaddle-gpu安装
记录服务器上cuda和对应的paddlepaddle-gpu安装
文章目录
- 记录服务器上cuda和对应的paddlepaddle-gpu安装
- 解决的问题：
- 问题1：C++ Traceback错误
- 问题2：CUBLAS_STATUS_NOT_INITIALIZED错误
- 原因剖析：
- 安装流程：
解决的问题：
问题1：C++ Traceback错误
问题2：CUBLAS_STATUS_NOT_INITIALIZED错误
Traceback (most recent call last):
File "/disks/sdb/user_space/niemengying/pipeline2/2-pre_annotation.py", line 13, in <module>
output = pipeline.predict(
input=f"tangut_dense2/{each_page_image_file}"
)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddleocr/_pipelines/pp_structurev3.py", line 211, in predict
return list(
self.predict_iter(
...<32 lines>...
)
)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/pipelines/_parallel.py", line 129, in predict
yield from self._pipeline.predict(
...<3 lines>...
)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/pipelines/layout_parsing/pipeline_v2.py", line 1137, in predict
table_res_all = list(
self.table_recognition_pipeline(
...<14 lines>...
),
)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/pipelines/_parallel.py", line 129, in predict
yield from self._pipeline.predict(
...<3 lines>...
)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/pipelines/table_recognition/pipeline_v2.py", line 1343, in predict
self.predict_single_table_recognition_res(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
crop_img_info["img"],
^^^^^^^^^^^^^^^^^^^^^
...<6 lines>...
use_ocr_results_with_table_cells,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/pipelines/table_recognition/pipeline_v2.py", line 989, in predict_single_table_recognition_res
table_structure_pred = next(self.wired_table_rec_model(image_array))
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/models/base/predictor/base_predictor.py", line 211, in __call__
yield from self.apply(input, **kwargs)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/models/base/predictor/base_predictor.py", line 267, in apply
prediction = self.process(batch_data, **kwargs)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/models/table_structure_recognition/predictor.py", line 92, in process
batch_preds = self.infer(x=x)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/models/common/static_infer.py", line 287, in __call__
pred = self.infer(x)
File "/disks/sdb/user_space/niemengying/miniconda3/lib/python3.13/site-packages/paddlex/inference/models/common/static_infer.py", line 252, in __call__
self.predictor.run()
~~~~~~~~~~~~~~~~~~^^
OSError: In user code:
ExternalError: CUBLAS error(1).
[Hint: 'CUBLAS_STATUS_NOT_INITIALIZED'. The cuBLAS library was not initialized. This is usually caused by the lack of a prior cublasCreate() call, an error in the CUDA Runtime API called by the cuBLAS routine, or an error in the hardware setup. To correct: call cublasCreate() prior to the function call; and check that the hardware, an appropriate version of the driver, and the cuBLAS library are correctly installed. ] (at /paddle/paddle/phi/kernels/funcs/blas/blas_impl.cu.h:42)
[operator < pd_op.while > error]
原因剖析：
- cuda安装问题：
nvcc -V返回
XXX@inspur:~/cuda12$ nvcc -V
Command 'nvcc' not found, but can be installed with:
apt install nvidia-cuda-toolkit
Please ask your administrator.
- cuda和paddle-gpu版本冲突
cuda 12.4, paddlepaddle-gpu 3.1.0
安装流程：
- 查看可以安装的最高版本，
nvidia-smi。版本cuda-12.4
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.90.07 Driver Version: 550.90.07 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
- 查看linux服务器版本，
cat /proc/version(注意：有空格，直接复制粘贴)。版本-Ubuntu 22.04
Linux version 5.15.0-151-generic (buildd@lcy02-amd64-092) (gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0, GNU ld (GNU Binutils for Ubuntu) 2.38) #161-Ubuntu SMP Tue Jul 22 14:25:40 UTC 2025
- 下载cuda
CUDA Toolkit下载界面
因为最高cuda版本为12.4，又cuda的稳定版本有11.8，12.6，12.9，因此，安装11.8版本。
在Downloads界面，选择如下配置：
在服务器上输入下载命令：wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
下载完成后。服务器上显示如下文件：
4. 服务器安装cuda
注意：因为没有sudo权限，直接运行sh cuda_11.8.0_520.61.05_linux.run
（此步骤可能需要耗费一些时间，请耐心等待）
使用回车键进行选择
输入accept，再回车：
利用上下键与Enter勾选对话框，选定内容如下：
随后选择Options，与Enter
选择Toolkit Options:
利用上下键与Enter勾选对话框，选定内容如下：
随后选择Change Toolkit Install Path，与Enter
填写cuda安装路径，与Enter
进入自己建立的文件夹中（使用cd XXX/cuda11.8），使用pwd获取绝对路径。
选择Done，一路返回：
选择Install，进行下载：
出现以下内容，则安装完成
5. 安装测试，命令nvcc -V，返回如下：
6. 安装paddle-gpu
paddle-gpu网址，选择如下：
服务器输入命令：
python -m pip install paddlepaddle-gpu==3.1.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
至此，安装了cuda与对应的paddle-gpu，问题解决。
