Build docker image to deploy from databricks

Rafael Novello

unread,

Aug 30, 2019, 3:32:07 PM8/30/19

to mlflow-users

Hi guys!

I'm a new MLFlow user and I would like to deploy models on Kubernets. Is there any tutorial for that?

I'm using Databricks to run my experiments and train my models but I don't know how to deploy from that.

How the community deals with that?

Thanks a lot!

Regards!

Matei Zaharia

unread,

Sep 12, 2019, 6:42:12 PM9/12/19

to Rafael Novello, mlflow-users

Hi Rafael,

The build-docker command described here can build a Docker container from your model that you can deploy to Kubernetes: https://www.mlflow.org/docs/latest/cli.html#mlflow-models-build-docker. You can run this command on a server that has Docker and have it read models from your Databricks workspace by configuring your MLflow client to talk to the Databricks tracking server using the MLFLOW_TRACKING_URI environment variable as described here: https://docs.databricks.com/applications/mlflow/tracking.html#log-to-a-tracking-server-from-the-api-or-cli. Then it will find the models that are stored on Databricks and download them and package them as Docker containers.

Matei

--
You received this message because you are subscribed to the Google Groups "mlflow-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mlflow-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mlflow-users/72e7a673-0d3a-4567-98d5-ef00851a04b6%40googlegroups.com.