How to install TensorFlow on Linux?

Install System Dependencies

• Ensure that Python 3.6 to 3.10 is installed on your system. TensorFlow does not support earlier Python versions.

 

• Upgrade pip to the latest version to avoid any compatibility issues:

  pip install --upgrade pip
  

 

• Install virtualenv to create isolated Python environments:

  pip install virtualenv
  

Create a Virtual Environment

• Create a directory for your TensorFlow environment. Enter the directory and set up the virtual environment:

  mkdir tensorflow_env
  cd tensorflow_env
  virtualenv .venv
  

 

• Activate your virtual environment to ensure that your TensorFlow installation doesn't interfere with other projects:

  source .venv/bin/activate
  

Install TensorFlow

• With the virtual environment activated, install TensorFlow. You can choose between CPU and GPU support:

   

  • For CPU support only:

    pip install tensorflow
    

   

  • For GPU support, ensure your NVIDIA drivers are updated, and install the GPU version:

    pip install tensorflow-gpu
    

 

• Confirm that TensorFlow has been installed by starting a Python shell and importing TensorFlow:

  python
  >>> import tensorflow as tf
  >>> print(tf.__version__)
  

Set Up GPU Support

• If you installed the GPU version, ensure that CUDA and cuDNN are correctly installed on your system. Visit the official NVIDIA documentation for the installation process.

 

• Install the NVIDIA CUDA Toolkit and cuDNN with matching versions supported by TensorFlow. Consider setting environment variables to point to these installations:

  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
  export CUDA_HOME=/usr/local/cuda
  

Test Your Installation

• Check if TensorFlow can access the GPU:

  python -c "import tensorflow as tf; print('GPU', 'available' if tf.config.list_physical_devices('GPU') else 'not available')"
  

 

• Run a simple TensorFlow script to test both installation and functionality:

  python -c "import tensorflow as tf; mnist = tf.keras.datasets.mnist; (x_train, y_train), (x_test, y_test) = mnist.load_data(); print('Training data shape:', x_train.shape)"
  

Maintain Your TensorFlow Environment

• Keep TensorFlow updated to benefit from the latest features and patches. This is particularly important for staying compatible with CUDA and cuDNN updates.

  pip install --upgrade tensorflow
  

 

• Use version control for your virtual environment's `requirements.txt` file to document the TensorFlow version and other dependencies:

  pip freeze > requirements.txt