neural_compressor.utils.create_obj_from_config

Utility methods to create corresponding objects from configuration.

Module Contents

Functions

get_func_from_config(func_dict, cfg[, compose])	Get the function or the composed function from configuration.
get_preprocess(preprocesses, cfg[, compose])	Get the preprocess function from configuration.
get_metrics(metrics, cfg[, compose])	Get the metrics function from configuration.
get_postprocess(postprocesses, cfg[, compose])	Get the postprocess function from configuration.
get_algorithm(algorithms, cfg[, compose])	Get the algorithms from configuration.
create_dataset(framework, data_source, cfg_preprocess, ...)	Create the dataset from the data source.
create_dataloader(framework, dataloader_cfg)	Create the dataloader according to the framework.
create_eval_func(framework, dataloader, adaptor, metric)	The interface to create evaluate function from config.
create_train_func(framework, dataloader, adaptor, ...)	The interface to create train function from config.

neural_compressor.utils.create_obj_from_config.get_func_from_config(func_dict, cfg, compose=True)

Get the function or the composed function from configuration.

neural_compressor.utils.create_obj_from_config.get_preprocess(preprocesses, cfg, compose=True)

Get the preprocess function from configuration.

neural_compressor.utils.create_obj_from_config.get_metrics(metrics, cfg, compose=True)

Get the metrics function from configuration.

neural_compressor.utils.create_obj_from_config.get_postprocess(postprocesses, cfg, compose=True)

Get the postprocess function from configuration.

neural_compressor.utils.create_obj_from_config.get_algorithm(algorithms, cfg, compose=False)

Get the algorithms from configuration.

Parameters:

• algorithms – the algorithm management.

• cfg – a dict contain the algo name and use it or not.

• compose – compose all algo or not. Defaults to False.

Returns:

All open algos.

neural_compressor.utils.create_obj_from_config.create_dataset(framework, data_source, cfg_preprocess, cfg_filter)

Create the dataset from the data source.

neural_compressor.utils.create_obj_from_config.create_dataloader(framework, dataloader_cfg)

Create the dataloader according to the framework.

neural_compressor.utils.create_obj_from_config.create_eval_func(framework, dataloader, adaptor, metric, postprocess_cfg=None, iteration=-1, tensorboard=False, fp32_baseline=False)

The interface to create evaluate function from config.

Parameters:

• framework (str) – The string of framework.

• dataloader (common.DataLoader) – The object of common.DataLoader.

• adaptor (obj) – The object of adaptor.

• metric – The evaluation metric.

• postprocess_cfg – The postprocess configuration.

• iteration – The number of iterations to evaluate.

• tensorboard – Whether to use tensorboard.

• fp32_baseline – The fp32 baseline score.

Returns:

The constructed evaluation function

neural_compressor.utils.create_obj_from_config.create_train_func(framework, dataloader, adaptor, train_cfg, hooks=None, callbacks=None)

The interface to create train function from config.

Parameters:

• framework (str) – The string of framework.

• dataloader (common.DataLoader) – The object of common.DataLoader.

• adaptor (obj) – The object of adaptor.

• train_cfg (dict) – The dict of training related config.

• hooks (dict) – The dict of training hooks, supported keys are: on_epoch_begin, on_epoch_end, on_step_begin, on_step_end. Their values are functions to be executed in adaptor layer.

Returns:

The constructed train function