Upload Model Parameters to Cloud

The Upload Model Parameters to Cloud job allows you to take a trained model and store its training to a blob in cloud storage, such as Google Cloud Storage (GCS), Amazon S3 (S3), or Azure Blob Storage. See Create the Upload Model Parameters to Cloud job for more information on creating the job and deploying a model with it.

Create the Upload Model Parameters to Cloud job

The Upload Model Parameters to Cloud job lets you store a trained model’s training to a blob in cloud storage.This document describes the steps to create the Upload Model Parameters to Cloud job and deploy a model with it.For more information, see Upload model parameters to Cloud.

Deploy a model

First, deploy a Smart Answers supervised model.

Create the job

Sign in to Managed Fusion and click your application.
Click Collections > Jobs > Add+ > Upload Model Parameters to Cloud to create a new instance of that job.
In the job ID field, enter a value of upload-cloud or a value appropriate for your model.
In the Model Name field, locate the name of the deployed model.
In the Cloud Path field, enter the location to store the parameters blob. You will also need to provide the Kubernetes secret that provides the access details for the cloud storage path.
Run the job, then check the cloud storage location to verify that the parameter file has been saved. The exported file will also have the version number (based on epoch time) of the model appended to it.

Take note of the full cloud path; you will need it to deploy the job later.

GCS

Obtain the Google Cloud Storage credentials JSON file for accessing the storage bucket you intend to use for transferring models.
Add the credentials JSON file as a secret with the following command:

kubectl create secret generic model-cloud-secret --from-file=credentials.json -n [NAMESPACE]

S3

Create a JSON file with the following keys:

”region” is optional and will be set to us-east-1 if a value is not entered

{
   "key": [KEY],
   "secret": [SECRET],
   "region": [REGION]
}

Upload the JSON file as a secret in the cluster:

kubectl create secret generic model-cloud-secret --from-file=model.json -n [NAMESPACE]

The secret can now be referenced as model-cloud-secret in the job configuration.

Azure Blob Storage

Create a JSON file with the layout below:

{
   "key": [KEY],
   "secret": [SECRET]
}

Create the model-cloud-secret secret in the Kubernetes cluster:

kubectl create secret generic model-cloud-secret --from-file=model.json -n [NAMESPACE]

Deploy a new model with the Upload Model Parameters job

The new model can run in either the same Managed Fusion cluster or a different cluster.If you are planning to transfer a model to a different cluster, note that you will need to set up the Kubernetes secrets in that cluster as well, though the access for deploying can be limited to read-only.

Click to create new Create Seldon Core Model Deployment job and toggle Advanced.
Enter the full cloud path, including the version number, in the Cloud Path box.
Enter the Kubernetes secret that will be used to access the path in the Kubernetes Secret Name for Cloud Storage Access box.
For this example, you will be using the job to deploy a trained Smart Answers model, so use the fusion-question-answering-seldon:5.4.0 model image from the lucidworks repository in the Image Name box.

Run the job. Confirm that the model deploys into the cluster and is running.

The Create Seldon Core Model Deployment job can pull down the saved model blobs and use them in deploying a new instance of the model, allowing for easier transference and versioning of trained models for rollback and testing.

UI tour

Index data

Query data

Metrics and analytics

Improve your queries

Administration

Developer documentation

Machine learning

Neural Hybrid Search

Release notes

FAQs

Deploy a model

Create the job

GCS

S3

Azure Blob Storage

Deploy a new model with the Upload Model Parameters job