setup script in tensorrt:18.11-py3 requires cuda 9.0

AI & Deep Learning Libraries and SDKs TensorRT
#1

Hi,
I cannot get the current tensorrt container to work properly:
Running

docker run --rm -it nvcr.io/nvidia/tensorrt:18.11-py3 /bin/bash

asks me to execute

/opt/tensorrt/python/python_setup.sh

which fails in the line:

UFF_PATH="$(python -c 'import uff; print(uff.__path__[0])')"

stating that libcublas.so.9.0 cannot be found and hence that tensorflow cannot be imported.

It seems that the version of tensorflow-gpu which is installed in the script does not work with cuda 10.
Am I doing something wrong?

#2

@larki,

This is a known issue that we’re working on. I see it’s currently tagged to be fixed in the 18.12 container (which could change if issues are found). In the meantime, I recommend using the 18.09 container since 18.10 only installs the non-GPU version of tensorflow.