autocast.logging.wandb

autocast.logging.wandb#

Utility helpers for configuring Weights & Biases logging.

create_notebook_logger(project='autocast-notebooks', name=None, *, enabled=True, tags=None, watch=None)[source]#

Create a WandbLogger for notebook use with minimal configuration.

Parameters:

  • project (str) – W&B project name (default: “autocast-notebooks”)

  • name (str | None) – Run name (default: auto-generated by wandb)

  • enabled (bool) – Whether to enable wandb logging (default: True)

  • tags (list[str] | None) – Tags to attach to the run

  • watch (str | None) – Model watching mode: “gradients”, “parameters”, “all”, or None to disable

Returns:

Configured logger instance, or None if disabled watch_cfg : Watch configuration for use with maybe_watch_model, or

None if disabled

Return type:

logger

Examples

>>> logger, watch_cfg = create_notebook_logger(
...     project="my-project",
...     name="experiment-1",
...     tags=["notebook", "exploration"],
... )
>>> # Then pass to trainer: L.Trainer(..., logger=logger)
>>> # And optionally: maybe_watch_model(logger, model, watch_cfg)
create_wandb_logger(logging_cfg, *, experiment_name, run_name=None, job_type=None, work_dir=None, config=None)[source]#

Instantiate a WandbLogger when enabled via the Hydra logging config.

Parameters:

  • logging_cfg (dict[str, Any] | DictConfig | None)

  • experiment_name (str)

  • run_name (str | None)

  • job_type (str | None)

  • work_dir (Path | None)

  • config (dict[str, Any] | DictConfig | None)

Return type:

tuple[WandbLogger | None, _WatchConfig | None]

log_metrics(logger, metrics, step=None)[source]#

Log scalar metrics when a trainer is not driving WandbLogger.

Parameters:
Return type:

None

maybe_watch_model(logger, model, watch_cfg)[source]#

Attach gradient/parameter watching if requested by the config.

Parameters:
Return type:

None