Runtime is a Docker image with a predefined infrastructure. It implements a set of specific methods that are used as an endpoints to the model. It’s responsible for running user-defined models. When you create a new application you also have to provide a corresponding runtime to each models’ instances.

We’ve already implemented a few runtimes which you can use in your own projects. They are all open-source and you can look up code if you need.

Framework Runtime Links
Python hydrosphere/serving-runtime-python Docker Hub
Tensorflow hydrosphere/serving-runtime-tensorflow Docker Hub
Spark hydrosphere/serving-runtime-spark Docker Hub

Note: If you are using a framework for which runtime isn’t implemented yet, you can open an issue in our Github.

What’s Next?