neural_compressor.experimental.nas.dynas

DyNAS approach class.

Module Contents

Classes

DyNAS

DyNAS approach.

class neural_compressor.experimental.nas.dynas.DyNAS(conf_fname_or_obj)

Bases: neural_compressor.experimental.nas.nas.NASBase

DyNAS approach.

Defining the pipeline for DyNAS approach.

Parameters:

conf_fname_or_obj (string or obj) – The path to the YAML configuration file or the object of NASConfig.

estimate(individual)

Estimate performance of the model.

Returns:

Evaluated metrics of the model.

init_for_search()

Initialize the search configuration.

search()

Execute the search process.

Returns:

Best model architectures found in the search process.

select_model_arch()

Select the model architecture.

create_acc_predictor()

Create the accuracy predictor.

create_macs_predictor()

Create the MACs predictor.

create_latency_predictor()

Create the latency predictor.

init_cfg(conf_fname_or_obj)

Initialize the configuration.