What is RocketML

RocketML is machine learning platform for distributed computing on the cloud. Data scientists can effortlessly build and train machine learning (ML) models using CPU/GPU clusters and deploy ML models as Docker containers with a REST API endpoint.

The RocketML stack consists of four layers:

1. Cloud Orchestrator: interacts with underlying cloud infrastructure to coordinate different cloud resources and manage end-to-end machine learning application life-cycle. These include Virtual Machines (VMs), Storage, Databases, IAM policies, Docker and Kubernetes resources.

2. High Performance Computing Backend: allows interactive creation and management of multi-node CPU and GPU clusters where high throughput and tightly coupled workloads can be executed in parallel.

3. MLOps Framework: facilitates several user tasks. For instance, it (i) creates and organizes user experiments (code, data, configuration and results); (ii) allows data science code to be packaged in a format reproducible on different platforms; (iii) deploys machine learning models in diverse serving environments, as well as stores, annotates, discovers, and manages models in a central repository.

4. Research Workbench: provides development tools such as JupyterLab, VSCode, and RStudio for coding in Python and R. The workbench comes pre-installed with several open-source libraries for a wide range of workloads.