Introducing InfinStor Starter with MLflow

InfinStor Starter with MLflow is an AWS hosted mlflow service. All the features of mlflow — MLflow Tracking, MLflow Projects, MLflow Models and Model Registry are available, ready to use with no setup, in the InfinStor Starter service.

Scalable, Serverless, Cloud Native Architecture

Implemented as Lambdas, with persistent information stored in DynamoDB

InfinStor Starter is architected using DynamoDB as the backend store and AWS Lambda as the processing engine. The advantages of this architecture are:

1. Cloud Scale: One user or thousands of users. A handful of models or thousands of versions of hundreds of models
2. No 24x7 mysql instances: Use of DynamoDB as the backend ensures that there are no always-on mysql instances running up costs on AWS
3. No Application Server Instances: Lambda is the processing engine for InfinStor Starter; hence there is no need to have 24x7 mlflow server instances running

Authenticated Service

Cognito User Pools, or federated to other OpenID, Active Directory

InfinStor Starter utilizes AWS Cognito for authentication. This provides a great deal of flexibility in managing users. The simplest is to treat InfinStor Starter users as an independent User Pool. More sophisticated user management can be accomplished by federating to other OAuth2 services or corporate Active Directory. All methods of access to the service — Web UI, Web UI embedded in Jupyterlab, CLI and API are protected by the authentication system.

Secure Service

All communication with the InfinStor service uses https connections with the highest recommended security setting — TLS 1.2

MLflow Projects with Cloud Single VM Execution

Run MLflow Projects in an EC2 Instance

InfinStor Starter supports MLflow Projects with a dedicated backend for running projects in a Single EC2 VM in the cloud. InfinStor Starter will automatically upload project files to a specified bucket in S3, start a VM of type specified by the user and run the MLflow project in the VM. The VM will wait 45 minutes for other tasks, and if no other MLflow projects are scheduled, the VM will automatically shut down.

MLflow Model Registry

InfinStor Starter includes a full implementation of the MLflow Model Registry using the specified Cloud Object Store as the store for models. Model versioning and artifact support, as defined by mlflow, are implemented in InfinStor Starter.

In summary, InfinStor Starter with MLflow is the best implementation of mlflow for Enterprise users. It is available as a multi-tenant multi-user SaaS in the cloud. It can also be operated in the customer’s AWS Account as a single-tenant multi-user SaaS. InfinStor Starter is perpetually free for use.

Please visit https://aws.amazon.com/marketplace/pp/B08FB3FYB8 to subscribe through AWS Marketplace. For operating this service in your own AWS Account in a single tenant multi-user mode, please contact sales@infinstor.com

Standard support is free and available by email — support@infinstor.com

Enterprise support is available for an extra charge. Please email sales@infinstor.com for details.

Service details are available at https://infinstor.com/mlflow and user documentation is available at https://docs.infinstor.com/