TensorFlow 2 Installation¶
Install Anaconda (Optional)¶
Create a new virtual env:
conda create -n env1 python=3.7
Remove an env:
conda remove --name env1 --all
List all envs:
conda env List
conda info --envs
List packages installed:
conda List
Install TensorFlow¶
Run:
pip install tensorflow
Verify installation (with GPU):
python -c "import tensorflow as tf; tf.test.is_gpu_available()"
GPU issue: if reporting missing library files, such as:
'Could not load dynamic library '...'; dlerror: ...'
Install:
Nvidia driver
CUDA Toolkit v10.1
CuDNN 7.6.5
CuDNN files unzipped into <INSTALL_PATH>/NVIDIA GPU Computing Toolkit/CUDA/v10.1/ (default: C:/Program Files/)
Environment setup:
‘Edit the system environment variables’
Click ‘Path’–>’Edit’
Add following paths:
<INSTALL_PATH>/NVIDIA GPU Computing Toolkit/CUDA/v10.1/bin <INSTALL_PATH>/NVIDIA GPU Computing Toolkit/CUDA/v10.1/libnvvp <INSTALL_PATH>/NVIDIA GPU Computing Toolkit/CUDA/v10.1/extras/CUPTI/libx64 # <INSTALL_PATH>/NVIDIA GPU Computing Toolkit/CUDA/v10.1/cuda/bin
For latest version (version >= 2.4.0), install:
Nvidia driver > 450.x
CUDA Toolkit v11
CuDNN 8.0.4
Optional: TensorRT 6.0
Issue:
dlerror: ‘cusolver64_10.dll’ not found
Solution:
Make a copy of ‘cusolver64_11.dll’ and rename as ‘cusolver64_10.dll’