"_CudaDeviceProperties" is already registered! when using paddleocr together with torch

There are a couple of issues with this workaround. However, no solution found. I am using paddlepaddle-gpu with cuda 12.2 and, when importing torch into the same environment, the error "_CudaDeviceProperties" is already registered! occurs. How to solve it?

Paddle version: 2.6.1 Paddle With CUDA: True

OS: Windows 10 GCC version: N/A Clang version: N/A CMake version: N/A Libc version: N/A Python version: 3.11.4

CUDA version: 12.2.128 Build cuda_12.2.r12.2/compiler.33053471_0 cuDNN version: N/A Nvidia driver version: 551.23 Nvidia driver List: GPU 0: NVIDIA GeForce RTX 4050 Laptop GPU

I also tried to downgrade to Cuda 12.0 and CUDNN 8.8 and the issue persists. It seems that can't coexist with Pytorch or Ultralytics package.

Additionally, this error is reported: (PreconditionNotMet) The third-party dynamic library (cublas64_118.dll;cublas64_11.dll) that Paddle depends on is not configured correctly. (error code is 126).

I had no issues with cpu version but with gpu seems there is no valid way to just install and run cuda version.