TensorFlow Serving for Kubeflow

This charm deploys TensorFlow Serving configured for use with Kubeflow to Kubernetes models in Juju.

Usage

Deploy an instance of this charm for each model you wish to serve, providing the path to the trained model via the model config option. It is also useful to name the deployed application after the model. For example, to deploy the demo inception model, you would use:

juju deploy cs:~johnsca/kubeflow-tf-serving inception --config model=gs://kubeflow-models/inception

Note that, while the Google Storage URL for the demo model is publicly accessible, it still requires Google credentials to access. If you used conjure-up canonical-kubernetes and deployed to GCP, this will be setup automatically for you.

If you are using Kubernetes on a different provider, you can also attach the model as a resource:

juju deploy cs:~johnsca/kubeflow-tf-serving inception --resource model=/path/to/model