CosmicAC Logo

Introduction

CosmicAC provides managed compute for machine learning and other compute-intensive workloads. Infrastructure setup can delay execution and shift focus away from development. CosmicAC handles infrastructure setup automatically, so jobs run immediately and scale as needed without manual server configuration.


Job types

A job is the unit of work in CosmicAC. Submitting a job triggers CosmicAC to provision resources, schedule the job, and manage the environment. CosmicAC offers two job types, each addressing a distinct compute use case.

GPU Container

A GPU Container job is an isolated KubeVirt virtual machine (VM) with direct access to GPU hardware. It is the most general-purpose job type in CosmicAC, suited for model training, experimentation, and custom workloads. The CLI connects you to the VM through an interactive shell session. The environment inside the VM is fully configurable. You can install packages, run custom scripts, and modify the operating system.

Managed Inference

A Managed Inference job serves open-source language models. It is suited for any use case that requires language model inference without managing serving infrastructure. CosmicAC provisions the model server and manages authentication and load balancing.


Why CosmicAC?

  • Minimal setup: You submit jobs through the CLI or web interface. CosmicAC provisions resources and schedules jobs automatically, with no manual server configuration.
  • Secure, isolated environments: Each job is a KubeVirt virtual machine, providing VM-level isolation with direct GPU access.
  • Fast provisioning: Jobs start in minutes. CosmicAC handles provisioning and scheduling automatically.
  • Built-in inference serving: CosmicAC handles authentication and load balancing for Managed Inference jobs.
  • Real-time notifications: CosmicAC sends email notifications when errors occur.

Who is CosmicAC for?

RoleWhy CosmicAC
ML EngineersTrain and experiment with models on on-demand GPU hardware
Data ScientistsRun and evaluate models without managing serving infrastructure
Software EngineersIntegrate language models into applications without infrastructure overhead
DevOps TeamsProvision and scale GPU resources automatically

Next steps

On this page