LightningCLI

class lightning.pytorch.cli.LightningCLI(model_class=None, datamodule_class=None, save_config_callback=<class 'lightning.pytorch.cli.SaveConfigCallback'>, save_config_kwargs=None, trainer_class=<class 'lightning.pytorch.trainer.trainer.Trainer'>, trainer_defaults=None, seed_everything_default=True, parser_kwargs=None, parser_class=<class 'lightning.pytorch.cli.LightningArgumentParser'>, subclass_mode_model=False, subclass_mode_data=False, args=None, run=True, auto_configure_optimizers=True, load_from_checkpoint_support=True)

Bases: object

Implementation of a configurable command line tool for pytorch-lightning.

Receives as input pytorch-lightning classes (or callables which return pytorch-lightning classes), which are called / instantiated using a parsed configuration file and / or command line args.

Parsing of configuration from environment variables can be enabled by setting parser_kwargs={"default_env": True}. A full configuration yaml would be parsed from PL_CONFIG if set. Individual settings are so parsed from variables named for example PL_TRAINER__MAX_EPOCHS.

For more info, read the CLI docs.

Parameters:

• model_class (Union[type[LightningModule], Callable[..., LightningModule], None]) – An optional LightningModule class to train on or a callable which returns a LightningModule instance when called. If None, you can pass a registered model with --model=MyModel.

• datamodule_class (Union[type[LightningDataModule], Callable[..., LightningDataModule], None]) – An optional LightningDataModule class or a callable which returns a LightningDataModule instance when called. If None, you can pass a registered datamodule with --data=MyDataModule.

• save_config_callback (Optional[type[SaveConfigCallback]]) – A callback class to save the config.

• save_config_kwargs (Optional[dict[str, Any]]) – Parameters that will be used to instantiate the save_config_callback.

• trainer_class (Union[type[Trainer], Callable[..., Trainer]]) – An optional subclass of the Trainer class or a callable which returns a Trainer instance when called.

• trainer_defaults (Optional[dict[str, Any]]) – Set to override Trainer defaults or add persistent callbacks. The callbacks added through this argument will not be configurable from a configuration file and will always be present for this particular CLI. Alternatively, configurable callbacks can be added as explained in the CLI docs.

• seed_everything_default (Union[bool, int]) – Number for the seed_everything() seed value. Set to True to automatically choose a seed value. Setting it to False will avoid calling seed_everything.

• parser_kwargs (Union[dict[str, Any], dict[str, dict[str, Any]], None]) – Additional arguments to instantiate each LightningArgumentParser.

• subclass_mode_model (bool) – Whether model can be any subclass of the given class.

• subclass_mode_data (bool) –

  Whether datamodule can be any subclass of the given class.

• args (Union[list[str], dict[str, Any], object, None]) – Arguments to parse. If None the arguments are taken from sys.argv. Command line style arguments can be given in a list. Alternatively, structured config options can be given in a dict or jsonargparse.Namespace.

• run (bool) – Whether subcommands should be added to run a Trainer method. If set to False, the trainer and model classes will be instantiated only.

• auto_configure_optimizers (bool) – Whether to automatically add default optimizer and lr_scheduler arguments.

• load_from_checkpoint_support (bool) – Whether save_hyperparameters should save the original parsed hyperparameters (instead of what __init__ receives), such that it is possible for load_from_checkpoint to correctly instantiate classes even when using complex nesting and dependency injection.

add_arguments_to_parser(parser)

Implement to add extra arguments to the parser or link arguments.

Parameters:

parser (LightningArgumentParser) – The parser object to which arguments can be added

Return type:

None

add_core_arguments_to_parser(parser)

Adds arguments from the core classes to the parser.

Return type:

None

add_default_arguments_to_parser(parser)

Adds default arguments to the parser.

Return type:

None

after_instantiate_classes()

Implement to run some code after instantiating the classes.

Return type:

None

before_instantiate_classes()

Implement to run some code before instantiating the classes.

Return type:

None

static configure_optimizers(lightning_module, optimizer, lr_scheduler=None)

Override to customize the configure_optimizers() method.

Parameters:

• lightning_module (LightningModule) – A reference to the model.

• optimizer (Optimizer) – The optimizer.

• lr_scheduler (Union[LRScheduler, ReduceLROnPlateau, None]) – The learning rate scheduler (if used).

Return type:

Any

init_parser(**kwargs)

Method that instantiates the argument parser.

Return type:

LightningArgumentParser

instantiate_classes()

Instantiates the classes and sets their attributes.

Return type:

None

instantiate_trainer(**kwargs)

Instantiates the trainer.

Parameters:

kwargs (Any) – Any custom trainer arguments.

Return type:

Trainer

static link_optimizers_and_lr_schedulers(parser)

Creates argument links for optimizers and learning rate schedulers that specified a link_to.

Return type:

None

parse_arguments(parser, args)

Parses command line arguments and stores it in self.config.

Return type:

None

setup_parser(add_subcommands, main_kwargs, subparser_kwargs)

Initialize and setup the parser, subcommands, and arguments.

Return type:

None

static subcommands()

Defines the list of available subcommands and the arguments to skip.

Return type:

dict[str, set[str]]