Make the KMP logs noverbose (Kaggle#1127)

See https://stackoverflow.com/questions/70250304/stop-tensorflow-from-printing-warning-message

Make the KMP logs noverbose (Kaggle#1127)

See https://stackoverflow.com/questions/70250304/stop-tensorflow-from-printing-warning-message

Add opencv classifier test (Kaggle#1119)

http://b/212703016

Add opencv classifier test (Kaggle#1119)

http://b/212703016

Upgrade to LightGBM 3.3.1 (Kaggle#1106)

http://b/207851889

Upgrade to LightGBM 3.3.1 (Kaggle#1106)

http://b/207851889

Remove /usr/local/cuda/compat from LD_LIBRARY_PATH (Kaggle#1092)

* Remove /usr/local/cuda/compat from LD_LIBRARY_PATH

http://b/203659565

* Add CUDA stubs to LD_LIBRARY_PATH to allow building on a CPU-only machine

* Add symlink for libcuda.so

Remove /usr/local/cuda/compat from LD_LIBRARY_PATH (Kaggle#1092)

* Remove /usr/local/cuda/compat from LD_LIBRARY_PATH

http://b/203659565

* Add CUDA stubs to LD_LIBRARY_PATH to allow building on a CPU-only machine

* Add symlink for libcuda.so

Add /opt/bin to GPU image PATH env variable (Kaggle#1089)

NVIDIA binaries from the host are mounted to /opt/bin.

This is how `nvidia-smi` is made accessible inside the user session docker container. See attached bugs for additional context.

b/202852505

Add /opt/bin to GPU image PATH env variable (Kaggle#1089)

NVIDIA binaries from the host are mounted to /opt/bin.

This is how `nvidia-smi` is made accessible inside the user session docker container. See attached bugs for additional context.

b/202852505