Documentation

Run Configuration

Set run configuration using the resnet_main() function.

Table 1 Model configuration function

Function

Description

Location

resnet_main()

Main function for run configuration, training, and validation.

official/r1/resnet/resnet_run_loop.py
  1. Import the following header files to the official/r1/resnet/resnet_run_loop.py file:
    1
2
    from npu_bridge.estimator.npu.npu_config import NPURunConfig 
from npu_bridge.estimator.npu.npu_estimator import NPUEstimator
  2. Replace Runconfig with NPURunconfig to configure run parameters.
    Tweak: resnet_main() in official/r1/resnet/resnet_run_loop.py
     1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
      ############## NPU modify begin #############
  # Replace Runconfig with NPURunconfig to adapt to the Ascend AI Processor. Save the checkpoint every 115200 steps and summary every 10000 times,
  # Preprocess data and enable the mixed precision mode to improve the training speed.
  run_config = NPURunConfig(
      model_dir=flags_obj.model_dir,
      session_config=session_config,
      save_checkpoints_steps=115200,
      enable_data_pre_proc=True,
      iterations_per_loop=100,
      # enable_auto_mix_precision=True,
      # Set precision_mode to allow_mix_precision.
      precision_mode='allow_mix_precision',
      hcom_parallel=True
  )
  ############## npu modify end ###############

    # The run configuration in the code is as follows.
  # run_config = tf.estimator.RunConfig(
  #     train_distribute=distribution_strategy,
  #     session_config=session_config,
  #     save_checkpoints_secs=60 * 60 * 24,
  #     save_checkpoints_steps=None)

    For details about how to set the mixed precision mode (precision_mode='allow_mix_precision' ), see Setting the Mixed Precision Mode.

  3. # Create NPUEstimator to replace tf.estimator.Estimator.

    Tweak: resnet_main() in official/r1/resnet/resnet_run_loop.py

     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
        # Replace tf.estimator.Estimator with NPUEstimator.
    classifier = NPUEstimator(
        model_fn=model_function, model_dir=flags_obj.model_dir, config=run_config,
        params={
            'resnet_size': int(flags_obj.resnet_size),
            'data_format': flags_obj.data_format,
            'batch_size': flags_obj.batch_size,
            'resnet_version': int(flags_obj.resnet_version),
            'loss_scale': flags_core.get_loss_scale(flags_obj,
                                                    default_for_fp16=128),
            'dtype': flags_core.get_tf_dtype(flags_obj),
            'fine_tune': flags_obj.fine_tune,
            'num_workers': num_workers,
            'num_gpus': flags_core.get_num_gpus(flags_obj),
        })
    # The creation of Estimator in the code is as follows.
    # classifier = tf.estimator.Estimator(
    #     model_fn=model_function, model_dir=flags_obj.model_dir, config=run_config,
    #     warm_start_from=warm_start_settings, params={
    #         'resnet_size': int(flags_obj.resnet_size),
    #         'data_format': flags_obj.data_format,
    #         'batch_size': flags_obj.batch_size,
    #         'resnet_version': int(flags_obj.resnet_version),
    #         'loss_scale': flags_core.get_loss_scale(flags_obj,
    #                                                 default_for_fp16=128),
    #         'dtype': flags_core.get_tf_dtype(flags_obj),
    #         'fine_tune': flags_obj.fine_tune,
    #         'num_workers': num_workers,
    #     })