Quickstart

This guide shows how to start a Studio, run a job, and stop it.

Start a Studio

from lightning_sdk import Studio

studio = Studio("my-studio")
studio.start()

Run a command

Use run to execute a shell command inside the Studio:

result = studio.run("echo 'Hello from Lightning!'")
print(result)

Run a job on a GPU machine

Pass a Machine to run on accelerated hardware:

from lightning_sdk import Studio, Machine

studio = Studio("my-studio")
job = studio.run("python train.py", machine=Machine.A10G)
job.wait()

Submit a named job

Use lightning_sdk.Job.run when you want to submit a named async job directly:

from lightning_sdk import Job, Machine

job = Job.run(
    name="train-model",
    command="python train.py",
    machine=Machine.A10G,
    studio="my-studio",
)
job.wait()

Submit multi-machine training

Use lightning_sdk.MMT.run for distributed training across multiple machines:

from lightning_sdk import MMT, Machine

mmt = MMT.run(
    name="distributed-train",
    command="python train.py",
    num_machines=4,
    machine=Machine.A10G,
    studio="my-studio",
)
mmt.wait()

Stop the Studio

studio.stop()

Next Steps

See the API Reference for the full public API.
Explore multi-machine training with lightning_sdk.MMT.run.
Deploy a model with Deployment.

Was this page helpful?