文档

NPUOutputTensorHook Constructor

Description

Constructs an object of class NPUOutputTensorHook. NPUOutputTensorHook is a hook for training, evaluation, and prediction of NPUEstimator, and it can call the user-defined output_fn every N step or at the end to print the output tensors. The NPUOutputTensorHook class inherits the LoggingTensorHook class and can call native APIs of the base class.

Prototype

def __init__(self, tensors,dependencies=None,output_fn=None,output_every_n_steps=0)

Options

Option

Input/Output

Description

tensors

Input

Name set of the input tensors, in dictionary or list format

dependencies

Input

Dependencies corresponding to tensors

output_fn

Input

Print function of tensor output

output_every_n_steps

Input

The user-defined output_fn, which is called when the session is executed for N times and the training script is executed.

Returns

An object of the NPUOutputTensorHook class

Restrictions

When Iterations_per_loop > 1, output_fn cannot be called as specified by output_every_n_steps.

Example

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
from npu_bridge.npu_init import *

# Define output_fn.
def output_fn(inputs):
  device_id = os.environ["ASCEND_DEVICE_ID"]
  ouput_file = os.path.join("/code", device_id, "test_npu_output_tensor.txt")
  for item in inputs:
    content = "step:{},loss:{}".format(str(item['global_step']), str(item['loss']))
    with open(ouput_file, 'a') as f:
      f.write(content)
      f.write("\n")

# Define output_hook for calling the user-defined output_fn.
        tensors = {'global_step': global_step, 'loss': loss}
        output_hook = NPUOutputTensorHook(
            tensors,
            dependencies=train_op_list,
            output_fn=output_fn,
            output_every_n_steps=10)
        train_hook.append(output_hook)

# Pass the hook to EstimatorSpec.
  return tf.estimator.EstimatorSpec(
      mode=mode,
      predictions=predictions,
      loss=loss,
      train_op=train_op,
      training_chief_hooks=train_hook,
      eval_metric_ops=metrics)