Platform

Segmind Platform includes two main components. Compute cluster to run your workloads, MLOps services and Datastore for long term data storage.

1) Compute

You can get access to compute via:

  1. IDE (JupyterLab & Visual Studio Code)
  2. Jobs

Cluster also runs other MLOps services including developer tools. If you are setting up Segmind for your enterprise or team, read more about creating a new cluster to learn what services run on your cluster.

2) Datastore

You can store your datasets and other file-based artefacts in Datastore.

3) Developer Tools

  1. Experiment tracking (ML Flow)
  2. Container Registry

Support for developer tools including Ray Tune HP tuning services are planned for future updates.

Concepts

Instance

Virtual computing environments that are running on a cluster with CPU, Memory and GPU as basic resources. Instance can be used to run Jupyter Lab, VS Code and Jobs Sessions. Instance state is always saved before stopping a session, except for Jobs.

Sessions

Session is a single stretch of an instance used for Jupyter Lab, VS Code or Jobs. Session state is saved to docker registry for Jupyter Lab and VS Code sessions.

Jupyter Lab Session

Integrated Development Environment or IDE to interact with an Instance. Jupyter Lab user interface supports a wide range of workflows in data science, scientific computing and machine learning.

VS Code Session

Integrated Development Environment or IDE to interact with an Instance. Visual Studio Code combines the simplicity of a code editor with what developers need for their core edit-build-debug cycle.

Jobs Session

You can submit and run a job via Segmind Python Library on an instance.

Storage volumes

Storage attached to instances. Volumes provide high-speed (High IOPS) throughput to run your workloads. You will lose your storage when you delete an instance. Use Datastore to store your datasets and other files that need to be stored for longer.

Datastore

Datastore is used to store your datasets and other static files such as checkpoint files.

Spot Instance

Instances that can be terminated at any time without notice. Spot instances are much cheaper (~ 50%) than normal instances and can be used to run non-critical workloads.

Base Image

Docker image used to create an instance. You can create a new base image by updating an existing one or by importing an image from a public or private docker registry.

Tracking

MLflow tracking service running on your cluster to record and compare parameters and results. You can use the Python Library to log your experiments.

For self-hosted (Segmind admin)

Cloud Service Provider (CSP)

You can choose your IaaS (AWS, GCP, Azure). Segmind will create your cluster used to run your workloads and other MLOps services.

Cluster

Segmind cluster is a set of computation resources and cloud services that your team can use to run their workloads and other MLOps related services. Segmind uses Kubernetes to create and manage compute resources.

Cluster Region

You can choose which region to run your workloads in. They differ based on CSP.