o 9wi0 @s(ddlZddlZddlZddlmZmZddlmZddlm Z m Z m Z m Z m Z mZmZmZmZmZmZddlZddlZddlmZddlmZddlmZddlmZdd lmZddl m!Z"dd l#m$Z$dd l m%Z%m&Z&m'Z'm(Z(m)Z)dd l*m+Z+dd l,m-Z-ddl.m/Z/ddl0m1Z1edZ2e2rddl3Z3ddl4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:e9ee:ddn e;e<d<e;e<d<edZ=Gdddej>j?j@Z@ee@fZAeee@fZBeeeee@fZCeee eDe eDe fe7fZEe e gefZFe egeee@ffZGGddde6ZHGddde%ZIGdddZJded eDfd!d"ZK d8ded#eee7e eDe ffd e eDe ffd$d%ZLded e e7ge eDe fffd&d'ZMd(ee ee d)ffd*e eDe fd e fd+d,ZNd-e;d eeDfd.d/ZOd0ee ePfd ePfd1d2ZQGd3d4d4ZRdee=d5e eDe fd e=fd6d7ZSdS)9N)partialupdate_wrapper) MethodType) AnyCallableDictIterableListOptionalSetTupleTypeTypeVarUnion)RequirementCache)_warn) Optimizer) LRScheduler)override)get_filesystem)CallbackLightningDataModuleLightningModuleTrainerseed_everything)_given_hyperparameters_context)MisconfigurationException) is_overridden)rank_zero_warnz jsonargparse[signatures]>=4.27.7)ActionConfigFileArgumentParser Namespaceclass_from_function"register_unresolvable_import_pathsset_config_read_modeT)fsspec_enabledr r! ModuleTypec s2eZdZdededededdf fdd ZZS) ReduceLROnPlateau optimizermonitorargskwargsreturnNcs$tj|g|Ri|||_dSN)super__init__r))selfr(r)r*r+ __class__R/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/lightning/pytorch/cli.pyr/<s zReduceLROnPlateau.__init__)__name__ __module__ __qualname__rstrrr/ __classcell__r3r3r1r4r';s*r'c s eZdZdZdddddededed ed ed d f fd dZ  d!dee dee e e e ffee ee ee ee fdededed eef ddZefddfdeeeeeedffdeded d fddZeddfdeeeedffdeded d fdd ZZS)"LightningArgumentParserzAExtension of jsonargparse's ArgumentParser for pytorch-lightning.z#Lightning Trainer command line toolPLF description env_prefix default_envr*r=r>r?r+r,Ncs>tstttj||||d|g|_i|_i|_dS)aInitialize argument parser that supports configuration file input. For full details of accepted arguments see `ArgumentParser.__init__ `_. Args: description: Description of the tool shown when running ``--help``. env_prefix: Prefix for environment variables. Set ``default_env=True`` to enable env parsing. default_env: Whether to parse environment variables. r<N)"_JSONARGPARSE_SIGNATURES_AVAILABLEModuleNotFoundErrorr.r/ callback_keys _optimizers_lr_schedulers)r0r=r>r?r*r+r1r3r4r/Ps   z LightningArgumentParser.__init__Tlightning_class. nested_key subclass_moderequiredcCst|r t|ts t|}t|tr?t|ttttfr?t|tr&|j ||r1|j ||d|dS|j ||dt|t ddSt d|d)aAdds arguments from a lightning class to a nested key of the parser. Args: lightning_class: A callable or any subclass of {Trainer, LightningModule, LightningDataModule, Callback}. nested_key: Name of the nested namespace to store arguments. subclass_mode: Whether allow any subclass of the given class. required: Whether the argument group is required. Returns: A list with the names of the class arguments added. F) fail_untypedrHT)rI instantiate sub_configszCannot add arguments from: zt. You should provide either a callable or a subclass of: Trainer, LightningModule, LightningDataModule, or Callback.)callable isinstancetyper" issubclassrrrrrBappendadd_subclass_argumentsadd_class_argumentsr)r0rErFrGrHr3r3r4add_lightning_class_argsks&      z0LightningArgumentParser.add_lightning_class_argsr( AUTOMATICoptimizer_classlink_tocCt|trtdd|DsJnt|tsJdddhd}t|tr/|j||fi|n |j||fddi|||f|j|<dS) adAdds arguments from an optimizer class to a nested key of the parser. Args: optimizer_class: Any subclass of :class:`torch.optim.Optimizer`. Use tuple to allow subclasses. nested_key: Name of the nested namespace to store arguments. link_to: Dot notation of a parser key to set arguments or AUTOMATIC. cs|]}t|tVqdSr-)rOr.0or3r3r4 z=LightningArgumentParser.add_optimizer_args..FparamsrJrIskiprKTN)rMtupleallrOrrQrRrC)r0rUrFrVr+r3r3r4add_optimizer_argss  z*LightningArgumentParser.add_optimizer_args lr_schedulerlr_scheduler_classcCrW) aAdds arguments from a learning rate scheduler class to a nested key of the parser. Args: lr_scheduler_class: Any subclass of ``torch.optim.lr_scheduler.{_LRScheduler, ReduceLROnPlateau}``. Use tuple to allow subclasses. nested_key: Name of the nested namespace to store arguments. link_to: Dot notation of a parser key to set arguments or AUTOMATIC. csrXr-)rOLRSchedulerTypeTuplerYr3r3r4r\r]z@LightningArgumentParser.add_lr_scheduler_args..Fr(r_rKTN)rMrarbrOrfrQrRrD)r0rerFrVr+r3r3r4add_lr_scheduler_argss  z-LightningArgumentParser.add_lr_scheduler_args)FT)r5r6r7__doc__rr8boolr/rrrrrrr r rSrr rcrfLRSchedulerTypergr9r3r3r1r4r:Ms|%    1 r:c@sxeZdZdZ    ddedededed ed ed d fd dZe de de ded d fddZ de de ded d fddZ d S)SaveConfigCallbackafSaves a LightningCLI config to the log_dir when training starts. Args: parser: The parser object used to parse the configuration. config: The parsed configuration that will be saved. config_filename: Filename for the config file. overwrite: Whether to overwrite an existing config file. multifile: When input is multiple config files, saved config preserves this structure. save_to_log_dir: Whether to save the config to the log_dir. Raises: RuntimeError: If the config file already exists in the directory to avoid overwriting a previous run config.yamlFTparserconfigconfig_filename overwrite multifilesave_to_log_dirr,NcCsJ||_||_||_||_||_||_d|_|s!td|ts#t ddSdS)NF save_configz`save_to_log_dir=False` only makes sense when subclassing SaveConfigCallback to implement `save_config` and it is desired to disable the standard behavior of saving to log_dir.) rmrnrorprqrr already_savedrrk ValueError)r0rmrnrorprqrrr3r3r4r/s zSaveConfigCallback.__init__trainer pl_modulestagecCs|jrdS|jrV|j}|dusJtj||j}t|}|js>|j r(| |nd}|j |}|r>t |jjd|d|j rV|j|dd|jj|j|d|j|jd|j rc||||d|_|j |j|_dS)NFz expected a to NOT exist. Aborting to avoid overwriting results of a previous run. You can delete the previous config file, set `LightningCLI(save_config_callback=None)` to disable config saving, or set `LightningCLI(save_config_kwargs={"overwrite": True})` to overwrite the config file.T)exist_ok) skip_nonerprq)rtrrlog_dirospathjoinrorrpis_global_zeroisfilestrategy broadcast RuntimeErrorr2r5makedirsrmsavernrqrs)r0rvrwrxr{ config_pathfs file_existsr3r3r4setups.  zSaveConfigCallback.setupcCdS)a Implement to save the config in some other place additional to the standard log_dir. Example: def save_config(self, trainer, pl_module, stage): if isinstance(trainer.logger, Logger): config = self.parser.dump(self.config, skip_none=False) # Required for proper reproducibility trainer.logger.log_hyperparams({"config": config}) Note: This method is only called on rank zero. This allows to implement a custom save config without having to worry about ranks or race conditions. Since it only runs on rank zero, any collective call will make the process hang waiting for a broadcast. If you need to make collective calls, implement the setup method instead. Nr3)r0rvrwrxr3r3r4rszSaveConfigCallback.save_config)rlFFT)r5r6r7rhr:r!r8rir/rrrrrsr3r3r3r4rks0 'rkc@s(eZdZdZddededdddddddf deeee e de ffdeeee e de ffdeeed ee e efd eeee deffd ee e efd eeefd eee e efe e e e efffdedededededdfddZd e e efdee e efe e efffddZdedefddZdede e efde e efddfddZd eddfd!d"Zd eddfd#d$Zd eddfd%d&Zd eddfd'd(Zede e ee ffd)d*Zd ededdfd+d,Zd-ed.e dedefd/d0Z ed eddfd1d2Z!d ededdfd3d4Z"dTd5d6Z#dTd7d8Z$dTd9d:Z%dedefd;d<Z&d=e e efd>e'e(defd?d@Z)d.ee defdAdBZ*e dUdCe dDe+dEee,defdFdGZ-d.ee ddfdHdIZ.dUd=e/dJe dKeedefdLdMZ0d.e ddfdNdOZ1d.e de e effdPdQZ2dTdRdSZ3dS)V LightningCLIzIImplementation of a configurable command line tool for pytorch-lightning.NTF model_class.datamodule_classsave_config_callbacksave_config_kwargs trainer_classtrainer_defaultsseed_everything_default parser_kwargssubclass_mode_modelsubclass_mode_datar*runauto_configure_optimizersr,cCs||_|pi|_||_|pi|_||_|pi|_| |_||_|p!t|_ |dup(| |_ ||_ |p0t |_ |dup7| |_||j\}}|| ||||j| | rV|jdnd|_|||||jdurv||jdSdS)a 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 :ref:`the CLI docs `. Args: model_class: An optional :class:`~lightning.pytorch.core.LightningModule` class to train on or a callable which returns a :class:`~lightning.pytorch.core.LightningModule` instance when called. If ``None``, you can pass a registered model with ``--model=MyModel``. datamodule_class: An optional :class:`~lightning.pytorch.core.datamodule.LightningDataModule` class or a callable which returns a :class:`~lightning.pytorch.core.datamodule.LightningDataModule` instance when called. If ``None``, you can pass a registered datamodule with ``--data=MyDataModule``. save_config_callback: A callback class to save the config. save_config_kwargs: Parameters that will be used to instantiate the save_config_callback. trainer_class: An optional subclass of the :class:`~lightning.pytorch.trainer.trainer.Trainer` class or a callable which returns a :class:`~lightning.pytorch.trainer.trainer.Trainer` instance when called. trainer_defaults: 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 :ref:`the CLI docs `. seed_everything_default: Number for the :func:`~lightning.fabric.utilities.seed.seed_everything` seed value. Set to True to automatically choose a seed value. Setting it to False will avoid calling ``seed_everything``. parser_kwargs: Additional arguments to instantiate each ``LightningArgumentParser``. subclass_mode_model: Whether model can be any `subclass `_ of the given class. subclass_mode_data: Whether datamodule can be any `subclass `_ of the given class. args: 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: Whether subcommands should be added to run a :class:`~lightning.pytorch.trainer.trainer.Trainer` method. If set to ``False``, the trainer and model classes will be instantiated only. N subcommand)rrrrrrrrr _model_classrrr_datamodule_classr_setup_parser_kwargs setup_parserparse_argumentsrmrnr _set_seed_add_instantiatorsbefore_instantiate_classesinstantiate_classes_run_subcommand)r0rrrrrrrrrrr*rr main_kwargssubparser_kwargsr3r3r4r/2s09      zLightningCLI.__init__cs@|fdd|D}fdd|D}||fS)Ncsi|] \}}|vr||qSr3r3rZkvsubcommand_namesr3r4 z5LightningCLI._setup_parser_kwargs..csi|] \}}|vr||qSr3r3rrr3r4rr) subcommandskeysitems)r0rrrr3rr4rs z!LightningCLI._setup_parser_kwargsr+cKs:|ddtjgtdi|}|jddtdd|S) z-Method that instantiates the argument parser. dump_headerzlightning.pytorch==z-cz--configz4Path to a configuration file in json or yaml format.)actionhelpNr3) setdefaultpl __version__r: add_argumentr)r0r+rmr3r3r4 init_parsers zLightningCLI.init_parseradd_subcommandsrrcCsD|jdi||_|ri|_|j|jfi|dS||jdS)z.model)rGNdataF)rGrH) rSrrr set_defaultsrrrrr)r0rmrr3r3r4add_core_arguments_to_parsers    z)LightningCLI.add_core_arguments_to_parsercCsT|||||||jr#|js|tf|js#|t | |dSr-) rradd_arguments_to_parserrrCrcrrDrgrf!link_optimizers_and_lr_schedulersrr3r3r4rs     zLightningCLI._add_argumentscCr)zImplement to add extra arguments to the parser or link arguments. Args: parser: The parser object to which arguments can be added Nr3rr3r3r4rrz$LightningCLI.add_arguments_to_parsercCshdhdhdhddS)zDDefines the list of available subcommands and the arguments to skip.>r datamoduleval_dataloaderstrain_dataloaders>rr dataloaders)fitvalidatetestpredictr3r3r3r3r4rs zLightningCLI.subcommandsc Ksi|_|}t|jtr|jnt|j}|D].}t||}t|}| |i}| d||j ||fi|} | |j|<|j || |dqdS)z%Adds subcommands to the input parser.r=)rN) _subcommand_parsersrrMrrNr"rgetattr_get_short_descriptiongetr_prepare_subcommand_parseradd_subcommand) r0rmr+parser_subcommandsrrfnr=rsubcommand_parserr3r3r4rs     zLightningCLI._add_subcommandsklassrcKsH|jdi|}||t||}|j|||d}||j|<|S)N)r`r3)rrsetradd_method_argumentsr)r0rrr+rmr`addedr3r3r4rs   z'LightningCLI._prepare_subcommand_parsercCsdi|j|j}|D]#\}\}}|dkrq t|tr#|||q t|}|j|||dq dS)z`Creates argument links for optimizers and learning rate schedulers that specified a ``link_to``.rT) compute_fnN)rCrDrrMralink_arguments_add_class_path_generator)rmoptimizers_and_lr_schedulerskey class_typerVadd_class_pathr3r3r4rs z.LightningCLI.link_optimizers_and_lr_schedulerscCsd|durttjdkrtdtjddd|dt|ttfr*|||_dS| ||_dS)z?Parses command line arguments and stores it in ``self.config``.NzLightningCLI's args parameter is intended to run from within Python like if it were from the command line. To prevent mistakes it is not recommended to provide both args and command line arguments, got: sys.argv[1:]=z, args=.) lensysargvrrMdictr! parse_objectrn parse_args)r0rmr*r3r3r4r s zLightningCLI.parse_argumentscCs~t|jj|jddd|_d|jvr|j|jj|_|jjt|ddt |j |j d|jjt|ddt |j |j ddS)NF)skip_link_targetsrzrrclir) subclassesr)yaml safe_loadrmdumprn config_dumpradd_instantiator_InstantiatorFn_get_module_typerrrrr0r3r3r4rs    zLightningCLI._add_instantiatorscCr)z:sz4LightningCLI.instantiate_trainer..rv)r)_parserrrBrr_instantiate_trainer)r0r+extra_callbackstrainer_configr3rr4r3s z LightningCLI.instantiate_trainerrnrcCsd}||vri||durg||<nt||ts||g||<|||||jvr@|j|}||t|tr;|n|g7<|jrh|ddsh|j||j|jt |j|jfi|j }|| |n t d|j jd|d|j di|S)Nr fast_dev_runFzThe `z` class does not expose the `z(` argument so they will not be included.r3)rMlistextendrrrrrrnr8rrPrrr7)r0rnrrvalueconfig_callbackr3r3r4r>s.      z!LightningCLI._instantiate_trainercCs|dur|jS|j|Sr-)rmrr0rr3r3r4rWs zLightningCLI._parserlightning_moduler(rdcCs4|dur|St|tr|||jddS|g|gfS)aOverride to customize the :meth:`~lightning.pytorch.core.LightningModule.configure_optimizers` method. Args: lightning_module: A reference to the model. optimizer: The optimizer. lr_scheduler: The learning rate scheduler (if used). N) schedulerr)r(rd)rMr'r))rr(rdr3r3r4configure_optimizers]s    z!LightningCLI.configure_optimizersc Cs|jsdS||}dttttdffdtttttttdfftffdttfdd}|t|j }|t |j }t |dkrCdSt |d ksOt |d kr^t d |jjd ||d |j |dd}||j|d}t|tszt||}|s~dSd}|r|j |dd} ||j|d}t| tst| |}td |jrtd |jjjd|jjdt|j|} |rt| |nd} t|j| | d} t| |jt| |j|j_dS)zOverrides the model's :meth:`~lightning.pytorch.core.LightningModule.configure_optimizers` method if a single optimizer and optionally a scheduler argument groups are added to the parser as 'AUTOMATIC'.Nr.registerr,csVg}|D]"\}\}}t|ts|f}|dkr(tfdd|Dr(||q|S)NrTc3s|]}t|VqdSr-)rOrrr3r4r\r]z`LightningCLI._add_configure_optimizers_method_to_model..get_automatic..)rrMraanyrP)rr  automaticr base_classrVr3r r4 get_automaticzs  zMLightningCLI._add_configure_optimizers_method_to_model..get_automaticrr`z{.add_configure_optimizers_method_to_model` expects at most one optimizer and one lr_scheduler to be 'AUTOMATIC', but found z. In this case the user is expected to link the argument groups and implement `configure_optimizers`, see https://lightning.ai/docs/pytorch/stable/common/lightning_cli.html#optimizers-and-learning-rate-schedulersr z..configure_optimizers` will be overridden by `z.configure_optimizers`.r )rrrr r rr8r rrCrfrDrrr2r5rrrMra_global_add_class_pathrrrinstantiate_class parametersrr rr) r0rrmr optimizers lr_schedulersrUoptimizer_initlr_scheduler_initrer(rdrr3r3r4rrsV "         z6LightningCLI._add_configure_optimizers_method_to_modelrrcCs|t|j|||S)zAUtility to get a config value which might be inside a subcommand.)rr8r)r0rnrrr3r3r4rszLightningCLI._getcCsxt|d|d}t|r|t|j|}t|||}||}|di|t|d|d}t|r:|dSdS)zRun the chosen subcommand.before_Nafter_r3)rrLrv_prepare_subcommand_kwargs)r0r before_fnrr fn_kwargsafter_fnr3r3r4rs    zLightningCLI._run_subcommandcs@fddjD}j|d<jdurj|d<|S)z@Prepares the keyword arguments to pass to the subcommand to run.cs$i|]\}}|jvr||qSr3)rrrr3r4rsz;LightningCLI._prepare_subcommand_kwargs..rNr)rrrr)r0rrr3rr4rs     z'LightningCLI._prepare_subcommand_kwargscCsb||jd}|dur dS|durtdd}nt|dd}|jr*||j|jd<dS||jd<dS)zSets the seed.rFNT)workers)rrnrr)r0 config_seedr3r3r4rs  zLightningCLI._set_seed)r,Nr-)4r5r6r7rhrkrr rr rrrrr8rrirArgsTyper/r rr:rrrrrr staticmethodr rrrrrrrrrr rrrrLRSchedulerTypeUnionr rr!rrrrr3r3r3r4r/s  &      2Z           "   > rrr,cCs|jd|jS)Nr)r6r5r r3r3r4_class_path_from_classsr% init_argscCs$t|tr |}t||pidS)N) class_pathr&)rMr!as_dictr%)rr&r3r3r4rs rcs"dtdtttfffdd }|S)Nr&r,cs t|Sr-)r)r&r r3r4rs z1_add_class_path_generator..add_class_path)r!rr8r)rrr3r r4rsrr*.initcCsV|di}t|ts|f}|ddd\}}t||gd}t||}||i|S)zInstantiates a class with the given args and init. Args: args: Positional arguments required for instantiation. init: Dict of the form {"class_path":...,"init_args":...}. Returns: The instantiated class object. r&r'rr)fromlist)rrMrarsplit __import__r)r*r)r+ class_module class_namemodule args_classr3r3r4rs  r componentc Csd|jdurdSz t|j}|jWSttjfy1}ztd|d|WYd}~dSd}~ww)NzFailed parsing docstring for z: )rhdocstring_parserparseshort_descriptionru ParseErrorr)r1 docstringexr3r3r4rs  "rrcCs"t|rt|tst|jS|Sr-)rLrMrNinspect signaturereturn_annotation)rr3r3r4rs rc@s@eZdZdededdfddZdeeded edefd d Z dS) rrrr,NcCs||_||_dSr-r)r0rrr3r3r4r/s z_InstantiatorFn.__init__rr*r+cOs||jj|ji}d|vrd|di|di|di}t|dd||i|WdS1s7wYdS)Nr' _class_pathr& dict_kwargsz(lightning.pytorch.cli.instantiate_module)hparams instantiator)rrrrr)r0rr*r+r=r3r3r4__call__s   $z_InstantiatorFn.__call__) r5r6r7rr8r/r r&rr?r3r3r3r4rs"rrncCsntdd}d|vr |j|ddd|ddd|Dd}n|j|ddd|d|i}||}|jS) NF) exit_on_errorr;r/)rIcSsi|] \}}|dkr||qS)r;r3rr3r3r4r1rz&instantiate_module..)r'r<)r rQrrRrrr/)rrnrmcfgr)r3r3r4instantiate_module+s  rBr-)Tr8r|r functoolsrrtypesrtypingrrrrr r r r r rrtorchr lightning_utilities.core.importsr"lightning_utilities.core.rank_zeror torch.optimrtorch.optim.lr_schedulerrtyping_extensionsrlightning.pytorchpytorchr#lightning.fabric.utilities.cloud_iorrrrrr+lightning.pytorch.core.mixins.hparams_mixinr&lightning.pytorch.utilities.exceptionsr)lightning.pytorch.utilities.model_helpersr%lightning.pytorch.utilities.rank_zerorr@r2 jsonargparserr r!r"r#r$objectlocalsr&optimrdr'rfr$rjr8r"OptimizerCallableLRSchedulerCallabler:rkrr%rrrrrNrrrBr3r3r3r4st  4                a1  $. &