composer.loggers.logger_hparams

Logger Hyperparameter classes.

Hparams

These classes are used with yahp for YAML-based configuration.

FileLoggerHparams	FileLogger hyperparameters.
InMemoryLoggerHparams	InMemoryLogger hyperparameters.
LoggerCallbackHparams	Base class for logger callback hyperparameters.
TQDMLoggerHparams	TQDMLogger hyperparameters.
WandBLoggerHparams	WandBLogger hyperparameters.

class composer.loggers.logger_hparams.FileLoggerHparams(log_level=LogLevel.EPOCH, filename='stdout', buffer_size=1, flush_interval=100, log_interval=1)

Bases: composer.loggers.logger_hparams.LoggerCallbackHparams

FileLogger hyperparameters.

See FileLogger for documentation.

Parameters

• filename (str, optional) – See FileLogger

• buffer_size (int, optional) – See FileLogger.

• log_level (LogLevel, optional) – See FileLogger.

• log_interval (int, optional) – See FileLogger.

• flush_interval (int, optional) – See FileLogger.

class composer.loggers.logger_hparams.InMemoryLoggerHparams(log_level=LogLevel.BATCH)

Bases: composer.loggers.logger_hparams.LoggerCallbackHparams

InMemoryLogger hyperparameters.

Parameters

log_level (str or LogLevel, optional) – See InMemoryLogger.

class composer.loggers.logger_hparams.LoggerCallbackHparams

Bases: yahp.hparams.Hparams, abc.ABC

Base class for logger callback hyperparameters.

Logger parameters that are added to TrainerHparams (e.g. via YAML or the CLI) are initialized in the training loop.

abstract initialize_object(config=None)

Initializes the logger.

Parameters

config (dict) – The configuration used by the trainer. The logger can optionally save this configuration.

class composer.loggers.logger_hparams.TQDMLoggerHparams

Bases: composer.loggers.logger_hparams.LoggerCallbackHparams

TQDMLogger hyperparameters. This class takes no parameters.

class composer.loggers.logger_hparams.WandBLoggerHparams(project=None, group=None, name=None, entity=None, tags='', log_artifacts=False, log_artifacts_every_n_batches=100, rank_zero_only=True, extra_init_params=<factory>, flatten_hparams=False)

Bases: composer.loggers.logger_hparams.LoggerCallbackHparams

WandBLogger hyperparameters.

Parameters

• project (str, optional) – WandB project name.

• group (str, optional) – WandB group name.

• name (str, optional) – WandB run name.

• entity (str, optional) – WandB entity name.

• tags (str, optional) – WandB tags, comma-separated.

• log_artifacts (bool, optional) – See WandBLogger.

• log_artifacts_every_n_batches (int, optional) – WandBLogger.

• extra_init_params (JSON Dictionary, optional) – See WandBLogger.

initialize_object(config=None)

Initializes the logger.

The config is flattened and stored as wandb.run.config. The list of algorithms in the config are appended to wandb.run.tags.

Parameters

config (Optional[Dict[str, Any]], optional) – The configuration used by the trainer.

Returns

WandBLogger – An instance of WandBLogger.