extensible.extensions.soleil#

Functions

init_from_checkpoint(checkpoint_file[, ...])

Instantiates the object (presumbaly a train manager) specified in the solconf description within a checkpoint file and modifies its checkpoint loader to load weights and fixtures from the specified checkpoint_file.

Classes

SoleilCheckpointSaver(*args[, root_config, ...])

Checkpoint saver for soleil-configured train managers.
class extensible.extensions.soleil.SoleilCheckpointSaver(*args, root_config: Path | str | None = None, package_overrides: List[str] | None = None, **kwargs)#

Bases: CheckpointSaver

Checkpoint saver for soleil-configured train managers. This class is assumed to be instantiated inside a *.solconf file in order to infer values for root_config and package_overrides. Otherwise, you need to supply values for these parameters.

Train managers can be loaded from saved checkpoints using init_from_checkpoint().

Warning

In order for init_from_checkpoint() to work, it is assumed that the solconf package where this class is instantiated resolves to a TrainManager instance – i.e., its as_type member instantiates a TrainManager object.

Parameters:

  • path – The path where all values will be saved.

  • saved_fixtures – The fixtures to save. By default, only 'epoch_num' is saved.

  • saved_attribs – If not specified, all train manager attributes with a state_dict attribute will be saved. This usually includes the model, optimizer and learning rate scheduler.

a checkpoint to load. When training, set to True to continue training if any checkpoints exist (training will fail with the default value False if checkpoints exist).

post_train_step_epoch(train_manager, fixtures, epoch_num)#

Save a checkpoint

extensible.extensions.soleil.init_from_checkpoint(checkpoint_file, overrides=None, solconf_package_dir=None)#

Instantiates the object (presumbaly a train manager) specified in the solconf description within a checkpoint file and modifies its checkpoint loader to load weights and fixtures from the specified checkpoint_file.

Note

Since any fixtures need to be loaded in the correct context (e.g., the train or eval stage), the weights and fixtures are not immediately loaded from the checkpoint. Rather the loaded object’s CheckpointLoader is modified by setting its ckpt_spec attribute to checkpoint_file. This will allow any fixtures (along with the weights) to load at the correct stage.

If needed, weights can be loaded explicitly on the returned train manager using CheckpointLoader.load_weights_and_fixtures, setting fixtures=None to skip loading fixtures.

Parameters:

  • overrides – Any overrides to apply when building the train manager from its solconf description.

  • solconf_package_dir – A directory where the checkpoint’s solconf package can be stored. By default, a temporary directory that is discarded before the function exists.