gigl.experimental.knowledge_graph_embedding.lib.checkpoint

Attributes

logger

Functions

maybe_load_checkpoint(model, optimizer, ...)	Load the model and optimizer checkpoints if they exist.
maybe_save_checkpoint(model, optimizer, ...[, ...])	Save the model and optimizer checkpoints if specified in the training configuration.

Module Contents

gigl.experimental.knowledge_graph_embedding.lib.checkpoint.maybe_load_checkpoint(model, optimizer, checkpointing_config)

Load the model and optimizer checkpoints if they exist.

Parameters:

• model (torch.nn.Module) – The model to load the checkpoint into.

• optimizer (torch.optim.Optimizer) – The optimizer to load the checkpoint into.

• checkpointing_config (gigl.experimental.knowledge_graph_embedding.lib.config.training.CheckpointingConfig) – The training configuration containing the checkpointing paths.

Returns:

True if the model and optimizer were loaded successfully, False otherwise.

Return type:

bool

gigl.experimental.knowledge_graph_embedding.lib.checkpoint.maybe_save_checkpoint(model, optimizer, checkpointing_config, checkpoint_id='')

Save the model and optimizer checkpoints if specified in the training configuration.

Parameters:

• model (torch.nn.Module) – The model to save the checkpoint for.

• optimizer (torch.optim.Optimizer) – The optimizer to save the checkpoint for.

• checkpointing_config (gigl.experimental.knowledge_graph_embedding.lib.config.training.CheckpointingConfig) – The training configuration containing the checkpointing paths.

• checkpoint_id (str) – An optional identifier for the checkpoint, used to differentiate between checkpoints if needed.

Returns:

The URI where the checkpoint was saved, or a Future object if saved asynchronously. If no checkpointing path is specified, returns None.

Return type:

Optional[Union[Future[Uri], Uri]]

gigl.experimental.knowledge_graph_embedding.lib.checkpoint.logger