Would be great if tensorflow_model_server opened by default an HTTP REST endpoint for better interoperability.
https://github.com/kubeflow/kubeflow/tree/master/components/k8s-model-server/http-proxy
There is one here...
and if you are using k8s, you can deploy like this https://github.com/yupbank/cmle_proxy/blob/master/app.yaml
This is also now supported natively.
Most helpful comment
This is also now supported natively.