NPUEstimatorSpec Constructor

Applicability

Product	Supported (√/x)
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	X
Atlas inference product	X
Atlas training product	√

Description

Constructor of the NPUEstimatorSpec class. The NPUEstimatorSpec class inherits the EstimatorSpec class of the TensorFlow and can call the native APIs of the base class to define specific model objects.

EstimatorSpec is the return data structure of model_fn, including the mode, predictions, loss, train_op, and export_outputs fields. If EstimatorSpec cannot meet the training requirements, define NPUEstimatorSpec to replace EstimatorSpec.

Prototype

class NPUEstimatorSpec(model_fn_lib.EstimatorSpec):
    def __new__(cls,
                mode,
                predictions=None,
                loss=None,
                train_op=None,
                eval_metric_ops=None,
                export_outputs=None,
                training_chief_hooks=None,
                training_hooks=None,
                scaffold=None,
                evaluation_hooks=None,
                prediction_hooks=None,
                host_call=None)

Parameters

Parameter	Input/Output	Description
mode	Input	Mode, indicating whether the current operation is training, validation, or inference. This parameter is inherited from EstimatorSpec. ModeKeys.TRAIN: training ModeKeys.EVAL: validation ModeKeys.PREDICT: inference
predictions	Input	Inference output tensor, required when mode is set to ModeKeys.PREDICT. It is a parameter inherited from EstimatorSpec.
loss	Input	Training loss. It is a parameter inherited from EstimatorSpec.
train_op	Input	Training operator. It is a parameter inherited from EstimatorSpec.
eval_metric_ops	Input	Dictionary of measurement results (based on tensor names). It is a parameter inherited from EstimatorSpec. The dictionary value can be one of the following: Metric instance. Result of calling the metric function, that is, the (metric_tensor, update_op) tuple.
export_outputs	Input	Saves a model and describes the output format of the model exported to SavedModel. This parameter is inherited from EstimatorSpec.
training_chief_hooks	Input	SessionRunHooks set of the primary node during training. It is a parameter inherited from EstimatorSpec.
training_hooks	Input	SessionRunHooks set during training. It is a parameter inherited from EstimatorSpec.
scaffold	Input	Scaffold definition (providing the capabilities of customizing saver, init_op, summary_op, and global_step). It is a parameter inherited from EstimatorSpec.
evaluation_hooks	Input	SessionRunHooks set during validation. It is a parameter inherited from EstimatorSpec.
prediction_hook	Input	SessionRunHooks set during inference. It is a parameter inherited from EstimatorSpec.
host_call	Input	Captures the summary information and sends the information of each step back to the host side. It is a new parameter in NPUEstimatorSpec. host_call is a tuple consisting of a function and a list or dictionary of tensors. It is used to return a list of tensors. host_call applies to train() and evaluate().

Returns

An object of the NPUEstimatorSpec class

Example

from npu_bridge.npu_init import *
...
host_call = (_host_call_fn, [global_step, loss])
return NPUEstimatorSpec(mode=tf.estimator.ModeKeys.TRAIN, loss=loss, train_op=train_op, host_call=host_call)

Parent topic: npu_bridge.estimator.npu.npu_estimator