save_model_config: Save and load model configuration as JSON

In keras3: R Interface to 'Keras'

Save and load model configuration as JSON

Description

Save and re-load models configurations as JSON. Note that the representation does not include the weights, only the architecture.

Usage

save_model_config(model, filepath = NULL, overwrite = FALSE)

load_model_config(filepath, custom_objects = NULL)

Arguments

model	Model object to save
filepath	path to json file with the model config.
overwrite	Whether we should overwrite any existing model configuration json at filepath, or instead ask the user via an interactive prompt.
custom_objects	Optional named list mapping names to custom classes or functions to be considered during deserialization.

Details

Note: save_model_config() serializes the model to JSON using serialize_keras_object(), not get_config(). serialize_keras_object() returns a superset of get_config(), with additional information needed to create the class object needed to restore the model. See example for how to extract the get_config() value from a saved model.

Value

This is called primarily for side effects. model is returned, invisibly, to enable usage with the pipe.

Example

model <- keras_model_sequential(input_shape = 10) |> layer_dense(10)
file <- tempfile("model-config-", fileext = ".json")
save_model_config(model, file)

# load a new model instance with the same architecture but different weights
model2 <- load_model_config(file)

stopifnot(exprs = {
  all.equal(get_config(model), get_config(model2))

  # To extract the `get_config()` value from a saved model config:
  all.equal(
      get_config(model),
      structure(jsonlite::read_json(file)$config,
                "__class__" = keras_model_sequential()$`__class__`)
  )
})

See Also

Other saving and loading functions:
export_savedmodel.keras.src.models.model.Model()
layer_tfsm()
load_model()
load_model_weights()
register_keras_serializable()
save_model()
save_model_weights()
with_custom_object_scope()

keras3 documentation built on June 8, 2025, 1:36 p.m.