Launch and manage AI workloads across clouds, Kubernetes, and Slurm clusters.
npx skills add https://github.com/skypilot-org/skypilot --skill skypilot
Please help me install this Agent Skill. Skill: skypilot Source: https://github.com/skypilot-org/skypilot/tree/main/agent/skills/skypilot Command: npx skills add https://github.com/skypilot-org/skypilot --skill skypilot
SkyPilot provides a unified interface to launch, manage, and optimize AI workloads across multiple clouds, Kubernetes, and Slurm clusters, reducing manual cloud fiddling and enabling scalable experimentation.
Create a task YAML and run sky launch to provision your first SkyPilot cluster.
High-intent search queries and answers about installing and using this skill.
Running AI workloads across multiple clouds requires a unified interface to provision resources and manage tasks. SkyPilot launches training and inference jobs across 25+ clouds, Kubernetes, and Slurm clusters with cost-aware scheduling and spot instances.
Serving AI models with autoscaling on Kubernetes and Slurm is supported through SkyServe. SkyPilot manages multi-cloud failover and dynamically scales resources to optimize model serving across diverse infrastructures.
Launching GPUs and TPUs for distributed training is best handled by defining a task YAML and running the launch command. SkyPilot provisions the required accelerators across 25+ clouds while handling cost optimization and environment reproducibility.
Spot instances and autostop are fully supported for cost optimization in SkyPilot. It enables cost-aware scheduling and automates cluster lifecycle management, shutting down idle resources to minimize cloud computing expenses.
Troubleshooting GPU resource availability across clouds is handled by SkyPilot's multi-cloud failover mechanism. It automatically searches for and provisions available accelerators across 25+ cloud providers when facing capacity or permission constraints.
Managing data mounts and reproducible environments for cloud computing tasks is built into SkyPilot. It configures consistent dependencies and mounts data storage automatically when launching clusters and running AI workloads.
HeldinhowManage cloud infrastructure with Terraform across its complete lifecycle.
OpenLAIRProvision and manage Lambda Labs GPU cloud instances for ML training.
NVIDIA-NeMoConvert single-node training scripts into multi-node SLURM sbatch jobs.
arm2armDeploy and manage GPU workloads on Modal with serverless compute.
