Skip to main content

​
What is SkyPilot?

SkyPilot is an open-source framework that simplifies running AI workloads on any cloud. It automatically finds the best price and availability across multiple cloud providers, provisions resources, and manages your jobs - all from a single unified interface.

​
Using Shadeform with SkyPilot

SkyPilot integrates with Shadeform to give you access to the best GPU prices across Shadeform’s marketplace of cloud providers. When you specify infra: shadeform in your SkyPilot task, SkyPilot will automatically search across all available providers in Shadeform to find the optimal GPU for your workload.

​
Setup

  1. Install SkyPilot 
    # Recommended: use a new conda env to avoid package conflicts.
# SkyPilot requires 3.7 <= python <= 3.13.
conda create -y -n sky python=3.10
conda activate sky

# Install just the Shadeform provider
pip install "skypilot[shadeform]"

# Optional: Install all supported providers
pip install "skypilot[all]"
    For more information on installing SkyPilot, see the SkyPilot documentation.
  2. Get your Shadeform API Key Sign up at shadeform.ai and obtain an API key from your account settings page.
  3. Configure your credentials In order to use Shadeform with SkyPilot you need to place your API key in a file called api_key in the ~/.shadeform directory on your machine. First, create the directory and then create the file with your API key. 
    mkdir ~/.shadeform
echo "your-api-key-here" > ~/.shadeform/api_key
  4. Verify the setup 
    sky check shadeform
You’re now ready to launch GPU workloads on Shadeform through SkyPilot!

​
Running Jobs

SkyPilot provides excellent documentation on how to get started with running jobs. See their quickstart guide and examples page. SkyPilot works by writing a YAML file that describes your job and then launching it with the sky launch command. A simple job that just prints “Hello, World!” would look like this: 
resources:
  infra: shadeform

run: |
  echo "Hello, World!"
If you save that file as job.yaml you can launch it with: 
sky launch job.yaml
For more information on what a cluster is, what job is, and how to build SkyPilot YAML jobs, refer to the quickstart guide and examples page.

​
Defining a Shadeform Resource

You can use Shadeform GPUs in three different ways in a SkyPilot YAML file by defining the infra, accelerators, or instance_type fields. To know what Shadeform GPUs are available, you can use the SkyPilot Catalog for Shadeform here. In the vms.csv file, you can see the available GPU types. InstanceType maps to the instance_type field in the SkyPilot YAML file. Accelerators maps to the accelerators field. AcceleratorCount shows how many GPUs that type has.

​
Find the cheapest available across all Shadeform providers

resources:
  infra: shadeform

​
Find the cheapest available for a specific GPU type

resources:
  infra: shadeform
  accelerators: A100:4 # this means A100 with 4 GPUs
Note: infra is optional if you define accelerators. But, when you have other SkyPilot integrations installed and you leave out infra, SkyPilot may pick your other integration to use if it has a cheaper accelerator of the same type available.

​
Find the cheapest available for a specific provider and GPU type

resources:
  infra: shadeform # optional when you define an instance_type
  accelerators: A100:4 # optional when you define an instance_type
  instance_type: hyperstack_A6000

​
Useful Commands

# Launch a job
sky launch task.yaml

# Launch with a specific cluster name
sky launch -c my-cluster task.yaml

# Check job status
sky status

# View logs
sky logs my-cluster

# Stop the cluster
sky stop my-cluster

# Terminate the cluster
sky down my-cluster