MXNet FAQ

.. toctree::
   :hidden:
   :glob:

   *

This section addresses common questions about how to use MXNet. These include performance issues, e.g., how to train with multiple GPUs. They also include workflow questions, e.g., how to visualize a neural network computation graph. These answers are fairly focused. For more didactic, self-contained introductions to neural networks and full working examples, visit the tutorials section.

API

• What's the difference between the Module and Gluon APIs for Python?

Modeling

• How do I fine-tune pre-trained models to a new dataset?

• How do I work with variable-length input in MXNet (bucketing)?

• How do I visualize neural networks as computation graphs?

Scale

• How can I train with multiple CPU/GPUs on a single machine with data parallelism?

• How can I train using multiple machines with data parallelism?

• How can I train using multiple GPUs with model parallelism?

Speed

• How do I use gradient compression with distributed training?

• Can I use nnpack to improve the CPU performance of MXNet?

• What are the best setup and data-handling tips and tricks for improving speed?

• How do I use mixed precision with MXNet or Gluon?

Deployment Environments

• Can I run MXNet on smart or mobile devices?

• How to use data from S3 for training?

• How to setup MXNet on AWS?

• How to do distributed training using MXNet on AWS?

• How do I run MXNet on a Raspberry Pi for computer vision?

• How do I run Keras 2 with MXNet backend?

• How to convert MXNet models to Apple CoreML format?

Security

• How to run MXNet securely?

Extend and Contribute to MXNet

• How do I join the MXNet development discussion?

• How do I contribute a patch to MXNet?

• How do I implement operators in MXNet backend?

• How do I create new operators in MXNet?

• How do I implement sparse operators in MXNet backend?

• How do I contribute an example or tutorial?

• How do I set MXNet's environmental variables?

Questions about Using MXNet

If you need help with using MXNet, have questions about applying it to a particular kind of problem, or have a discussion topic, please use our forum.

Issue Tracker

We track bugs and new feature requests in the MXNet Github repo in the issues folder: mxnet/issues.

Roadmap

MXNet is evolving fast. To see what's next and what we are working on internally, go to the MXNet Roadmap.