nvflare.app_opt.tf.params_converter module¶

class KerasModelToNumpyParamsConverter(supported_tasks: List[str] | None = None)[source]¶

This class converts the dict of {layer_name: layer_weights} to a flattened numpy dict.

The layer_name is from a keras `layer.name` The layer_weights is from `layer.get_weights()`

convert(params: Any, fl_ctx) → Any[source]¶: Flattens layer weights dict.

class NumpyToKerasModelParamsConverter(supported_tasks: List[str] | None = None)[source]¶

This class converts a flattened numpy dict back into a dict of {layer_name: layer_weights}.

The result can be used by `keras_model.get_layer(layer_name).set_weights(layer_weights)`

convert(params: Any, fl_ctx) → Any[source]¶: Unflattens layer weights dict.