Skip to content

mmts_cli.rst

Latest commit

 

History

History
57 lines (41 loc) · 1.45 KB

mmts_cli.rst

File metadata and controls

57 lines (41 loc) · 1.45 KB
Raw

MMT CLI examples

Use the CLI for multi-machine training when you want a scriptable distributed launch path without writing Python orchestration code.

Prerequisites

$ pip install lightning-sdk -U
$ lightning login
$ lightning studio start --name sdk-tutorial-studio --teamspace owner/teamspace --machine CPU --create

Run a Studio-backed MMT

$ lightning mmt run \
    --name sdk-tutorial-mmt \
    --teamspace owner/teamspace \
    --studio sdk-tutorial-studio \
    --num-machines 2 \
    --machine CPU \
    --env RUN_MODE=distributed \
    --command "python train_distributed.py --epochs 1"

Inspect and list MMTs

$ lightning mmt inspect sdk-tutorial-mmt --teamspace owner/teamspace
$ lightning mmt list --teamspace owner/teamspace --sort-by status

Run an image-backed MMT

$ lightning mmt run \
    --name sdk-image-mmt \
    --teamspace owner/teamspace \
    --image pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime \
    --num-machines 2 \
    --machine L4 \
    --command "python -m torch.distributed.run --nproc_per_node=1 train.py"

Clean up

$ lightning mmt stop sdk-tutorial-mmt --teamspace owner/teamspace
$ lightning mmt delete sdk-tutorial-mmt --teamspace owner/teamspace