8. Next steps

There are plenty of resources to help you progress with developing on IPUs.

8.1. Support

Support is included with your access. You can request support by clicking on the Submit a ticket link on the Graphcore Support portal.

For general help, discussions and announcements, please join our Graphcore Slack Community.

When looking for answers or asking questions on StackOverflow, use the tag “ipu”.

Note

If you want to use IPUs to test your own models or in a production environment, please contact us using the form on the Graphcore Contact page and we will help you to get set up.

8.2. Graphcore tutorials

There are Graphcore tutorials available in the Graphcore GitHub Examples repository. You will have to modify the advanced options for the the Gradient Notebook you create to specify the Workspace URL as “https://github.com/graphcore/examples/” and enter a suitable container from the Graphcore Docker hub into Container Name as described in Section 5.5, Advanced options (optional). Some tutorials that you might find interesting to run are:

• Porting a simple PyTorch model to the IPU

• PyTorch Geometric on IPUs at a glance

• Using half and mixed precision in PyTorch

8.3. Further reading

Graphcore’s documentation website contains useful documents for learning how to work with IPUs. Some examples are:

• The IPU Programmer’s Guide

• Switching from GPUs to IPUs for Machine Learning Models

• PyTorch for the IPU: User Guide