1 configs.wandb_config

configs.wandb_config

Weights & Biases logger configuration for Lightning experiments.

This module owns WandbConfig and its config-as-factory translation to Lightning’s WandbLogger, including local storage, run identity, grouping, resume, and checkpoint-artifact options. It does not define metric schemas, start training, or create a W&B run until WandbConfig.setup_target is called.

The config keeps W&B identity, resume, grouping, and artifact settings in the same config-as-factory tree as the rest of an ARIA-NBV experiment.

1.1 Classes

Name Description
WandbConfig Wrapper around Lightning’s WandbLogger.

1.1.1 WandbConfig

configs.wandb_config.WandbConfig()

Wrapper around Lightning’s WandbLogger.

1.1.1.1 References

https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.loggers.wandb.html

1.1.1.2 Attributes

Name Description
target_type Return Lightning’s W&B logger class for config-factory inspection.
name Optional display name for the run.
project Destination W&B project name.
entity Optional W&B user or team entity.
run_id Stable run id used for creation or resumption.
resume Policy for an already existing run_id.
anonymous Optional W&B anonymous-mode override.
offline Write events locally without contacting the W&B service.
log_model Lightning checkpoint-to-W&B artifact logging policy.
checkpoint_name Optional name for uploaded checkpoint artifacts.
tags Optional run tags used for filtering and comparison.
group Optional group shared by related runs or trials.
job_type Optional role label such as training, evaluation, or sweep.
prefix Optional namespace prepended to logged metric keys.

1.1.1.3 Methods

Name Description
init_kwargs Return the shared W&B run-identity arguments for non-Lightning reporters.
setup_target Instantiate a logger rooted in PathConfig’s W&B directory.
1.1.1.3.1 init_kwargs
configs.wandb_config.WandbConfig.init_kwargs()

Return the shared W&B run-identity arguments for non-Lightning reporters.

Lightning-only checkpoint and metric-prefix options deliberately remain exclusive to :meth:setup_target. This keeps small reporting tools on the same project, identity, grouping, and offline-mode contract without pretending that they are Lightning loggers.

1.1.1.3.2 setup_target
configs.wandb_config.WandbConfig.setup_target(**kwargs)

Instantiate a logger rooted in PathConfig’s W&B directory.

Back to top