cosmos

Manage Cosmos world model deployment, inference, and serverless training on Nebius.

17|8|Updated Apr 7, 2026
One-click install
npx skills add https://github.com/nebius/nebius-physical-ai --skill cosmos-nebius
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cosmos
Source: https://github.com/nebius/nebius-physical-ai/tree/main/skills/tools/cosmos
Command: npx skills add https://github.com/nebius/nebius-physical-ai --skill cosmos-nebius

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill addresses the complexity of managing world model serving, inference, and serverless training workflows, specifically for synthetic data and video generation tasks on Nebius infrastructure.

Core Features & Use Cases

  • Model Lifecycle Management: Deploy, serve, and infer using Cosmos world models with support for multiple backends.
  • Serverless Training: Execute smoke validation and fine-tuning jobs in a managed cloud environment.
  • Sim2Real Evaluation: Coordinate parallel sibling GPU jobs for VLM-based rollout evaluation using Reason models.

Quick Start

Use the cosmos skill to initiate a serverless training smoke test on the Nebius platform.

Frequently Asked Questions about cosmos

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I manage synthetic data generation using Cosmos world models?

To manage synthetic data generation with Cosmos world models, this Skill orchestrates deployment, serving, and inference lifecycles. It supports multiple backends and serverless training workflows directly on Nebius infrastructure.

How do I run serverless training jobs for world models on Nebius?

To run serverless training jobs for world models on Nebius, this Skill executes smoke validation and fine-tuning workloads. It manages the managed cloud environment required for these synthetic data generation tasks.

Do I need Nebius CLI and Hugging Face credentials to deploy Cosmos models?

Yes, deploying Cosmos models requires integration with the Nebius CLI for backend selection and configured Hugging Face credentials for model access. These prerequisites are necessary to initiate serverless training smoke tests.

How does VLM-based sim-to-real evaluation work for synthetic data?

VLM-based sim-to-real evaluation works by coordinating parallel sibling GPU jobs for rollout evaluation. It uses Reason models to assess the synthetic data generated by the Cosmos world model.

Can I execute smoke validation for world model training in a serverless environment?

Yes, you can execute smoke validation for world model training in a serverless environment. This Skill manages these validation workloads within a managed cloud environment on Nebius infrastructure.

What is the best way to evaluate synthetic video generation rollouts?

The best way to evaluate synthetic video generation rollouts is using parallel VLM-based sim-to-real evaluation workflows. This Skill coordinates these sibling GPU jobs using Reason models to validate the output.