Open dbarrau opened 2 years ago
Hey, @dbarrau , I experienced this (apparently some version incompatibility) and I solved it with the following
# Check libcudnn8 version
!apt-cache policy libcudnn8
# Install latest version
!apt install --allow-change-held-packages libcudnn8=8.4.1.50-1+cuda11.6
# Export env variables
!export PATH=/usr/local/cuda-11.4/bin${PATH:+:${PATH}}
!export LD_LIBRARY_PATH=/usr/local/cuda-11.4/lib64:$LD_LIBRARY_PATH
!export LD_LIBRARY_PATH=/usr/local/cuda-11.4/include:$LD_LIBRARY_PATH
!export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/extras/CUPTI/lib64
# Install tensorflow
!pip install tflite-model-maker==0.4.0
# !pip uninstall -y tensorflow && pip install -q tensorflow==2.9.1
!pip install pycocotools==2.0.4
!pip install opencv-python-headless==4.6.0.66
All steps were carried out according to the documentation, and when attempted to train the model after compiling, I received this error:
Something else I noted, after installing and importing
tensorflow_text
, I was required to restart the runtime, and then models with bidirectional LSTM layers would not be able to train. In this next case, the traceback error is:Anyone having these issues within Colab? Thanks
You can see the error accessing my colab, but it has lots there. It will take some time to run and reach to the latest part. https://colab.research.google.com/drive/1wRbbEeLDmD4gNjlWvCmxYtDxc8lliaNw?usp=sharing