Save the date! Google I/O returns May 18-20 Register now

Module: tf.saved_model

TensorFlow 2 version

Public API for tf.saved_model namespace.


builder module: SavedModel builder.

constants module: Constants for SavedModel save and restore operations.

experimental module: Public API for tf.saved_model.experimental namespace.

loader module: Loader functionality for SavedModel with hermetic, language-neutral exports.

main_op module: SavedModel main op.

signature_constants module: Signature constants for SavedModel save and restore operations.

signature_def_utils module: SignatureDef utility functions.

tag_constants module: Common tags used for graphs in SavedModel.

utils module: SavedModel utility functions.


class Builder: Builds the SavedModel protocol buffer and saves variables and assets.


build_signature_def(...): Utility function to build a SignatureDef protocol buffer.

build_tensor_info(...): Utility function to build TensorInfo proto from a Tensor. (deprecated)

classification_signature_def(...): Creates classification signature from given examples and predictions.

contains_saved_model(...): Checks whether the provided export directory could contain a SavedModel.

get_tensor_from_tensor_info(...): Returns the Tensor or CompositeTensor described by a TensorInfo proto. (deprecated)

is_valid_signature(...): Determine whether a SignatureDef can be served by TensorFlow Serving.

load(...): Loads the model from a SavedModel as specified by tags. (deprecated)

load_v2(...): Load a SavedModel from export_dir.

main_op_with_restore(...): Returns a main op to init variables, tables and restore the graph. (deprecated)

maybe_saved_model_directory(...): Checks whether the provided export directory could contain a SavedModel.

predict_signature_def(...): Creates prediction signature from given inputs and outputs.

regression_signature_def(...): Creates regression signature from given examples and predictions.

save(...): Exports the Trackable object obj to SavedModel format.

simple_save(...): Convenience function to build a SavedModel suitable for serving. (deprecated)

Other Members

  • ASSETS_DIRECTORY = 'assets'
  • ASSETS_KEY = 'saved_model_assets'
  • CLASSIFY_INPUTS = 'inputs'
  • CLASSIFY_METHOD_NAME = 'tensorflow/serving/classify'
  • GPU = 'gpu'
  • LEGACY_INIT_OP_KEY = 'legacy_init_op'
  • MAIN_OP_KEY = 'saved_model_main_op'
  • PREDICT_INPUTS = 'inputs'
  • PREDICT_METHOD_NAME = 'tensorflow/serving/predict'
  • PREDICT_OUTPUTS = 'outputs'
  • REGRESS_INPUTS = 'inputs'
  • REGRESS_METHOD_NAME = 'tensorflow/serving/regress'
  • REGRESS_OUTPUTS = 'outputs'
  • SAVED_MODEL_FILENAME_PB = 'saved_model.pb'
  • SAVED_MODEL_FILENAME_PBTXT = 'saved_model.pbtxt'
  • SERVING = 'serve'
  • TPU = 'tpu'
  • TRAINING = 'train'
  • VARIABLES_DIRECTORY = 'variables'
  • VARIABLES_FILENAME = 'variables'