Does this skill support model serving?

Yes, through Sky Serve, it provides capabilities for deploying model endpoints with features like readiness probes, replica policies, and automatic scaling based on traffic.

What happens if a spot instance is preempted during a job?

When using managed jobs via 'sky jobs launch', SkyPilot automatically detects preemption, finds new available resources, and restarts the job from the last checkpoint to ensure completion.

Can I use this skill for distributed multi-node training?

Yes, SkyPilot simplifies multi-node setup by handling network configuration, environment variables, and gang scheduling for frameworks like PyTorch and Ray.

Which cloud providers does SkyPilot support?

SkyPilot supports over 20 providers including major platforms like AWS, GCP, Azure, and Kubernetes, as well as specialized GPU providers like Lambda Cloud and RunPod.

How does SkyPilot reduce machine learning infrastructure costs?

It automatically identifies the cheapest available resources across different clouds and regions and enables the use of spot instances, which can offer 3-6x cost savings compared to on-demand instances.

SkyPilot Multi-Cloud Orchestration

Name: SkyPilot Multi-Cloud Orchestration
Author: Orchestra-Research

byOrchestra-Research

•

3,983

•

クラウドインフラストラクチャ

Automates ML workload deployment across multiple cloud providers with intelligent cost optimization and spot instance management.

This skill empowers Claude to orchestrate complex machine learning workflows across more than 20 cloud providers, including AWS, GCP, Azure, and Kubernetes. It specializes in reducing infrastructure overhead by automatically selecting the most cost-effective regions and leveraging spot instances with built-in auto-recovery. Ideal for AI researchers and engineers, it provides standardized patterns for distributed multi-node training, managed batch jobs, and scalable model serving while effectively eliminating vendor lock-in through a unified interface.

主な機能

01Distributed multi-node training support with gang scheduling

02Managed spot instance execution with automatic preemption recovery

03Production-grade model serving with autoscaling via Sky Serve

04Automatic cost-based cloud and region selection for GPU workloads

053,983 GitHub stars

06Unified orchestration for 20+ cloud providers and Kubernetes

ユースケース

01Running large-scale LLM fine-tuning across distributed GPU clusters

02Executing long-running batch processing jobs on low-cost spot instances

03Deploying scalable model inference endpoints with multi-cloud failover

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add orchestra-research/ai-research-skills skypilot

For use in Claude.ai and ChatGPT

Download Skill