0

I am currently doing an OCR technology project using only PaddleOCR and running only in GPU on Google Colab. Recently, Google Colab updated its cuDNN version from 7.6 to 8.1, which resulted in my training model being unable to work. The error message is shown below. May I know if there are any ways to downgrade the cuDNN version, or if there are any ways to solve this error issue?

C++ Traceback (most recent call last):

No stack trace in paddle, may be caused by external reasons.

Error Message Summary:

FatalError: Segmentation fault is detected by the operating system. [TimeInfo: *** Aborted at 1665370329 (unix time) try "date -d @1665370329" if you are using GNU date ***] [SignalInfo: *** SIGSEGV (@0x0) received by PID 523 (TID 0x7f4afdf67780) from PID 0 ***]

WhySoGud
  • 1

1 Answers1

0

I had the same problem (I think) and was able to fix by reinstalling the correct packages as specified on the paddle website here.

  1. Check google colab's CUDA version by either:

!nvcc --version or !nvidia-smi

  1. Go to https://www.paddlepaddle.org.cn/en and select the appropriate info, including the CUDA version. For me on 17OCT2022, it was version 11.2.

  2. Modify the install code provided to work on google colab as such:

!pip install paddlepaddle-gpu==2.3.1.post112 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html

  1. Confirm the paddlepaddle-gpu installed correctly by entering into a cell: import paddle paddle.utils.run_check()

(step #4 will NOT tell you if you installed the right version according to CUDA but it will at least give you some confidence that the packages were otherwise correct)

loni415
  • 1