model_to_npu_estimator

Applicability

Product	Supported
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	☓
Atlas inference products	☓
Atlas training products	√

Description

Converts the model constructed by using Keras to an NPUEstimator object.

Prototype

def model_to_npu_estimator(keras_model=None,
                           keras_model_path=None,
                           custom_objects=None,
                           model_dir=None,
                           checkpoint_format='saver',
                           config=None,
                           job_start_file='')

Parameters

Parameter	Description
keras_model	Built Keras model object. This parameter and keras_model_path cannot be input at the same time.
keras_model_path	Path for storing the built Keras model on the drive. You can use the save() method of the Keras model to generate a Keras model in HDF5 format. This parameter and keras_model cannot be input at the same time.
custom_objects	Dictionary of user-defined objects. If a user-defined layer or function is used during Keras construction, custom_objects must be used during model loading.
model_dir	Model directory, which is used to save or restore model files. If this parameter is not set, the value of model_dir in the config file is used. If both parameters are set, the values of the two parameters must be the same. If both parameters are set to None, the temporary directory /tmp is used.
checkpoint_format	Sets the format of the checkpoint file saved by NPUEstimator during training. saver (default): Save the model using tf.train.Saver(). checkpoint: Save the model using tf.train.Checkpoint (). Compared with tf.train.Saver, tf.train.Checkpoint supports delayed variable recovery in instant execution mode.
config	NPURunConfig class object, which is used to configure NPUEstimator running parameters. For details about the constructor of the NPURunConfig class, see NPURunConfig Constructor.
job_start_file	Path of the configuration file used to start the training process in the CSA scenario.

Returns

An NPUEstimator object is returned based on the input Keras model.

Restrictions

Currently, only the function model and sequence model (Keras graph construction mode) can be converted into an NPUEstimator object using the model_to_npu_estimator API.

Parent topic: npu_bridge.estimator.npu.keras_to_npu