Install Neuron Plugin for TensorBoard

The Neuron plugin for TensorBoard is available starting with Neuron v1.13.0.

To install the Neuron plugin, first enable ML framework Conda environment of your choice, by running one of the following:

  • Enable PyTorch-Neuron Conda enviroment:

source activate aws_neuron_pytorch_p36
  • Enable TensorFlow-Neuron Conda enviroment:

    source activate aws_neuron_tensorflow_p36
    
  • Enable MXNet-Neuron Conda enviroment:

    source activate aws_neuron_mxnet_p36
    

Then run the following:

If you are using the DLAMI TensorFlow-Neuron Conda environment, please run the following to update TensorBoard before installing the Neuron plugin.

pip install "tensorboard<=2.4.0" --force-reinstall

Modify Pip repository configurations to point to the Neuron repository:

tee $VIRTUAL_ENV/pip.conf > /dev/null <<EOF
[global]
extra-index-url = https://pip.repos.neuron.amazonaws.com
EOF
pip install tensorboard-plugin-neuron

Install Neuron TensorBoard (Deprecated)

Warning

TensorBoard-Neuron is deprecated and no longer compatible with Neuron tools version 1.5 and higher. Neuron tools version 1.5 is first introduced in Neuron v1.13.0 release. Please use the Neuron plugin for TensorBoard instead.

To install Tensorboard, first enable ML framework Conda environment of your choice, by running one of the following:

  • Enable PyTorch-Neuron Conda enviroment:

source activate aws_neuron_pytorch_p36
  • Enable TensorFlow-Neuron Conda enviroment:

    source activate aws_neuron_tensorflow_p36
    
  • Enable Neuron Conda enviroment for Neuron Apache MXNet (Incubating):

    source activate aws_neuron_mxnet_p36
    

Then run the following:

pip install tensorboard-neuron
  • Installing tensorflow-neuron<=1.15.5.1.2.9.0 will automatically install tensorboard-neuron as a dependency. The final version, 1.15.5.1.2.9.0, is part of Neuron v1.12.2 release.

  • To verify tensorboard-neuron is installed correctly, run tensorboard_neuron -h | grep run_neuron_profile. If nothing is shown, please retry installation with the --force-reinstall option.