Huawei uses machine translation combined with human proofreading to translate this document to different languages in order to help you better understand the content of this document.
Note: Even the most advanced machine translation cannot match the quality of professional translators.
Huawei shall not bear any responsibility for translation accuracy and it is recommended that you refer to the English document (a link for which has been provided).
Configuring Run Parameters
Run Parameters
Run parameters are configured using the resnet_main() function.
Table 7-5 Model configuration function
Function |
Description |
Location |
|---|---|---|
resnet_main() |
Main function for run configuration, training, and validation. |
/official/r1/resnet/resnet_run_loop.py |
- Import the following header files to the official/r1/resnet/resnet_run_loop.py file:
from npu_bridge.estimator.npu.npu_config import NPURunConfig from npu_bridge.estimator.npu.npu_estimator import NPUEstimator
- Runconfig is replaced with NPURunconfig to configure the run parameters and adapt to the Ascend AI Processor.Tweak: resnet_main() in official/r1/resnet/resnet_run_loop.py
############## 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 Mixed Precision.
- Create an NPUEstimator and replace tf.estimator.Estimator with NPUEstimator to adapt to the Ascend AI Processor.
Tweak: resnet_main() in official/r1/resnet/resnet_run_loop.py
############## npu modify begin ############# # Replace tf.estimator.Estimator with NPUEstimator to adapt to the Ascend AI Processor. 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), }) ############## npu modify end ############### # 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, # })
