Deep Learning AMI
Developer Guide

Keras with MXNet

This tutorial shows how to activate and use Keras 2 with the MXNet backend on a Deep Learning AMI with Conda.

Activate Keras with the MXNet backend and test it on the DLAMI with Conda

  1. To activate Keras with the MXNet backend, open an Amazon Elastic Compute Cloud (Amazon EC2) instance of the DLAMI with Conda.

    • For Python 3, run this command:

      $ source activate mxnet_p36
    • For Python 2, run this command:

      $ source activate mxnet_p27
  2. Start the iPython terminal:

    (mxnet_p36)$ ipython
  3. Test importing Keras with MXNet to verify that it is working properly:

    import keras as k

    The following should appear on your screen (possibly after a few warning messages).

    Using MXNet backend


    If you get an error, or if the TensorFlow backend is still being used, you need to update your Keras config manually. Edit the ~/.keras/keras.json file and change the backend setting to mxnet.

Keras-MXNet Multi-GPU Training Tutorial

Train a convolutional neural network (CNN)

  1. Open a terminal and SSH into your DLAMI.

  2. Navigate to the ~/examples/keras-mxnet/ folder.

  3. Run nvidia-smi in your terminal window to determine the number of available GPUs on your DLAMI. In the next step, you will run the script as-is if you have four GPUs.

  4. (Optional) Run the following command to open the script for editing.

    (mxnet_p36)$ vi
  5. (Optional) The script has the following line that defines the number of GPUs. Update it if necessary.

    model = multi_gpu_model(model, gpus=4)
  6. Now, run the training.

    (mxnet_p36)$ python


Keras-MXNet runs up to two times faster with the channels_first image_data_format set. To change to channels_first, edit your Keras config file (~/.keras/keras.json) and set the following: "image_data_format": "channels_first".

For more performance tuning techniques, see Keras-MXNet performance tuning guide.

More Info

  • You can find examples for Keras with a MXNet backend in the Deep Learning AMI with Conda ~/examples/keras-mxnet directory.

  • For even more tutorials and examples, see the Keras-MXNet GitHub project.