Initialize the current directory with a project

$ wandb init

Describe the next run using Markdown

$ wandb describe

Checkout the commit associated with the run and restore the config variables

$ wandb restore run-id

Pull the latest run from the cloud

$ wandb pull latest

Push extra files to a run

$ wandb push run-id model.json weights.h5

Manually login to Weights & Biases

$ wandb login

Get the status of the files in the current project

$ wandb status

List the runs in your project

$ wandb runs

List the projects in your account

$ wandb projects

W&B configuration makes tracking exactly what configuration parameters were used in a given training automatic. You can benefit from W&B configuration tracking by changing a single line of your training script. If you decide to read configuration from our configuration object we store defaults in a YAML file at the root of your project. The configuration object automatically looks for overrides in the environment (if WANDB_ is prepended to the name) as well as in command line flags.

Initialize a directory for configuration. This creates a file named config-defaults.yaml in the current directory.

$ wandb config init

Set configuration variables

$ wandb config set batch_size=25
$ wandb config set batch_size=25 -d "The size of a mini-batch"
$ wandb config set batch_size=25 epochs=10

Import configuration from existing code. If you're currently just setting parameters with optional comments in python, this is a great way to get started organizing parameters.

$ wandb config import

Pasting the following into the editor will set the appropriate values in the config.

# The size of a mini-batch
batch_size=25
epochs = 10

Remove a configuration variable

$ wandb config rm batch_size

Show the current configuration

$ wandb config show
$ wandb config show -f json

Setup the configuration and optionally a client. As long as you've run wandb init in the current directory there's no need to provide credentials or a project name.

import wandb
conf = wandb.Config()
wandb.sync(files=["*.h5"])

Set or use configuration parameters. This enables W&B to keep track of all configuration parameters for a given training run.

conf.some_rate = 1.5
if conf.some_rate < 1:
    pass

If your training code already uses something like argparse or tensorflow.flags you can pass the parsed object to get tracking for free.

args = parser.parse_known_args()
conf = wandb.Config(args)
conf = wandb.Config(FLAGS)

The next time you push a run, the configuration parameters will be synced to W&B.

Pull the run specified from the cloud. The default project will be overridden if a "/" is in the run name.

wandb.pull("cool/stuff")

Push the current project to the cloud. File paths are relative to the current working directory.

wandb.push("some-run", files=["model.json", "weights.h5"])