TensorFlow

Back to troubleshooting start

TF2 installation

Note

In some cases you need to install the latest pip version to avoid errors, and also to install the latest TensorFlow version instead of using the conda version.

upgrading pip

pip install --upgrade pip

installing TF2

pip install tensorflow

This is supposed to install the current stable release for CPU and GPU. But please note that you may need to install the version that comes with CUDA.

python3 -m pip install 'tensorflow[and-cuda]'

verifying TF2 installation

python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

TF2 and TensorRT fix:

Note

Run these commands after installing TF2 with its requirements to fix the Cuda errors and to make it see TensorRT

installing TensorRT

pip install nvidia-pyindex
pip install nvidia-tensorrt

Note

some cuda errors persist in TF2 but they are warrnings not a “real” MemoryError.

fixing Cuda errors

#for the error for tensorRT even after installing:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH: “/home/user/miniconda3/envs/tf/lib/python3.11/site-packages/tensorrt_libs/”

replace the path with your path to tensorrt_libs, and user with your username in the path.

Keras

Note

AttributeError: module ‘keras’ has no attribute ‘ops’

Keras ops either is in another tree, or I have to downgrade to another version

Tensorflow Profiler

Caution

bug(not showing the profile logs): it is a bug from the tensorboard/profiler you just need to move the logs from inside the train or validations directories inside the logs directory up one level so they are on the same directory level of the “train” and “validation” directories, not inside them, then the profile will show.

Solution source (Not English).

Discussion of the problem on Github.

Note

Please note that tensorboard profiler is getting updated constantly, this bug may not always appear, but other issues were reported with the profiler, so please research any problems faced, it may not be your installation or environment.