Skip to content
Get packs

SkyPilot Multi-Cloud GPU Orchestration

Run ML training and batch jobs across many clouds with spot-instance recovery and cost-aware GPU selection

Source Download skill
Repository
Orchestra-Research/AI-Research-SKILLs
Stars
14,195
Licence
MIT
Files
3 files

What it does

This skill teaches an AI agent how to use SkyPilot, an open-source tool that launches machine learning jobs on GPUs from many cloud providers through one configuration format. Once installed, the agent can write SkyPilot task files, pick the cheapest cloud or region for a GPU type, run jobs on cheaper spot (interruptible) instances with automatic recovery, set up multi-node distributed training, and deploy model serving endpoints. Actually launching anything requires your own cloud accounts and credentials, and the resulting cloud spend is yours.

Great for

  • ML engineers who train models on rented GPUs and want to avoid being locked into one cloud
  • Teams looking to cut GPU costs by using spot instances with checkpointing
  • Researchers who need to run multi-node training jobs without writing cluster setup by hand

Skip it if

  • People with no cloud accounts or who only need a local GPU; the skill is built around paid cloud resources and will fail without credentials
  • Unattended cost-sensitive use without budget limits: the agent can launch expensive multi-GPU clusters, and autostop or teardown must be set correctly to avoid ongoing charges
  • Production serving with strict uptime requirements, since spot instances can be preempted

Try asking

  • “Write a SkyPilot YAML that trains my model on 4 A100s using spot instances and resumes from checkpoints if preempted”
  • “Find the cheapest cloud and region to run an 8x H100 job and launch it”
  • “Set up a 2-node torchrun training job on SkyPilot and show me how to view its logs”
  • “Deploy a vLLM serving endpoint with autoscaling using SkyPilot Sky Serve”
Step-by-step guide: how to use SkyPilot Multi-Cloud GPU OrchestrationInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 3 original files, the MIT licence and an attribution note.

    skypilot.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip skypilot.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip skypilot.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Set autostop or use sky down when finished, and ask the agent to confirm the cloud, region and GPU count before launching, since costs accrue on running clusters
  • Use managed jobs (sky jobs launch) with checkpointing to Cloud storage when using spot instances, so training can resume after preemption
  • Review the generated YAML before launch, especially file_mounts paths and any commands in setup or run, because the skill's examples execute shell commands on remote machines

What's inside

  • references/advanced-usage.md
  • references/troubleshooting.md
  • SKILL.md

Redistributed unmodified from Orchestra-Research/AI-Research-SKILLs under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.