labml

Synopsis

labml [options] <command> [command_specific_args]

Options

-h, --help

Display usage summary.

Commands

app-server [*args*]

Start labml server.

Important

Requires labml-app to be installed. You can install it with pip install labml-app -U.

args:

--ip <ip_address>

IP address to bind the local server instance. If unspecified, it will bind to 0.0.0.0, which makes the server accessible through all the assigned IPs of the computer.

Default: 0.0.0.0

--port <port>

Port to use for the local server instance.

Default: 5005

-h, --help

Display usage summary.

capture [*args*]

Capture the output of any command or program as an experiment.

args: command to run. If no command is specified, data from the STDIN will be used; so you can pipe an output of a program to labml capture.

Note

For example, labml capture python train.py

launch [*args*]

Run a distributed training session with torch.distributed.launch.

args: command to run the distributed training session with torch.distributed.launch.

monitor

Monitor the hardware of the computer.

Important

Requires psutil to be installed.

Requires py3nvml to be installed to monitor GPUs.

service

Setup the hardware monitoring as a service and start monitoring.

Note

Run monitor first to make sure monitoring works without any issue, before setting up the service.

service-run

This is what gets called by the service installed usign labml service.

Note

This is for internal use and you shouldn’t run this manually.

dashboard

Open the dashboard to view experiments.

Deprecated since version 0.4.118: Use labml server instead.