composer.loggers.wandb_logger#

Log to Weights and Biases (https://wandb.ai/)

Classes

WandBLogger

Log to Weights and Biases (https://wandb.ai/)
class composer.loggers.wandb_logger.WandBLogger(log_artifacts=False, log_artifacts_every_n_batches=100, rank_zero_only=True, init_params=None)[source]#

Bases: composer.core.logging.base_backend.LoggerCallback

Log to Weights and Biases (https://wandb.ai/)

Parameters

  • log_artifacts (bool, optional) โ€“ Whether to log artifacts (Default: False).

  • log_artifacts_every_n_batches (int, optional) โ€“

    Interval at which to upload artifacts to wandb from the run_directory. Logging very frequently (e.g. on every batch) can substantially slow down training, so we recommend doing so infrequently. Only applicable when log_artifacts is True (default: 100).

  • rank_zero_only (bool, optional) โ€“

    Whether to log only on the rank-zero process. When logging artifacts, it is highly recommended to log on all ranks. Artifacts from ranks โ‰ฅ1 will not be stored, which may discard pertinent information. For example, when using Deepspeed ZeRO, it would be impossible to restore from checkpoints without artifacts from all ranks (default: False).

  • init_params (Dict[str, Any], optional) โ€“ Parameters to pass into wandb.init (see WandB documentation).