nvflare.app_common.abstract.model_persistor module

class ModelPersistor(filter_id: str | None = None)

Bases: LearnablePersistor, ABC

Abstract class. Implementations will need to implement the load_model() and save_model() methods to persist & load the current ModelLearnable.

Parameters:

filter_id – Optional string that defines a filter component that is applied to prepare the model to be saved, e.g. for serialization of custom Python objects.

get(model_file: str, fl_ctx: FLContext) → object

get_model(model_file: str, fl_ctx: FLContext) → ModelLearnable

Retrieve a specific model by file name.

This method is called by get() to load a specific model from the inventory. Persistors that support multiple models (e.g., for cross-site evaluation) should override this method to load the specified model file.

Simple persistors that only work with a single model (like NPModelPersistor) do not need to implement this method.

Parameters:

• model_file – Name or path of the model file to retrieve

• fl_ctx – FLContext

Returns:

ModelLearnable object containing the model data

Raises:

NotImplementedError – If the persistor doesn’t support retrieving specific models

get_model_inventory(fl_ctx: FLContext) → Dict[str, ModelDescriptor]

Get the model inventory of the ModelPersister.

Parameters:

fl_ctx – FLContext

Returns:

ModelDescriptor

Return type:

A dict of model_name

load(fl_ctx: FLContext) → ModelLearnable

Load the Learnable object.

Parameters:

fl_ctx – FLContext

Returns:

Learnable object loaded

abstract load_model(fl_ctx: FLContext) → ModelLearnable

Initialize and load the model.

Parameters:

fl_ctx – FLContext

Returns:

Model object

save(learnable: ModelLearnable, fl_ctx: FLContext)

Persist the Learnable object.

Parameters:

• learnable – the Learnable object to be saved

• fl_ctx – FLContext

abstract save_model(model: ModelLearnable, fl_ctx: FLContext)

Persist the model object.

Parameters:

• model – Model object to be saved

• fl_ctx – FLContext