job
launch
job
launch
Launch a Komodo job.
Usage
komo job launch [OPTIONS] CONFIG_FILE
Options
| Option | Required | Description | Type |
|---|---|---|---|
| ‑‑name | No | Job name. If not provided, a name will be generated. | str |
| ‑c, ‑‑cloud | No | Name of the cloud to use. If not provided, Komodo will select a cloud for you based on GPU prices and availability. | str |
| ‑g, ‑‑gpus | No | Which gpus to use, in the format <GPU type>:<Num GPUs> (eg. A100:4) | str |
| ‑n, ‑‑num-nodes | No | The number of nodes to launch for this job (for multi-node distributed training) | int |
| ‑d, ‑‑detach | No | Exit after the job is started instead of following the logs | flag |
| ‑r, ‑‑run | No | Command to run when the job starts | |
| ‑e, ‑‑env | No | Specify an environment variable in the form --env key=value or --env key (value will come from the value of $key in your shell). You can use this flag multiple times to specify multiple environment variables. | str |
Examples
Run the hello world job described in the quickstart
Run the same job, but this time specify a different command to run
Run a training job distributed over 4 nodes. Your job will have access to the following environment variables:
SKYPILOT_NODE_RANK: rank (an integer ID from 0 to num_nodes-1) of the node executing the task
SKYPILOT_NODE_IPS: a string of IP addresses of the nodes reserved to execute the task, where each line contains one IP address
SKYPILOT_NUM_GPUS_PER_NODE: number of GPUs reserved on each node to execute the task; the same as the count in accelerators: <name>:<count>
Run a job and exit as soon as the job has started