classification_flow Namespace Reference

Functions
def	get_parser ()
def	prepare_model (model_name, data_dir, per_channel_quantization, batch_size_train, batch_size_test, batch_size_onnx, calibrator, pretrained=True, ckpt_path=None, ckpt_url=None)
def	main (cmdline_args)
def	evaluate_onnx (onnx_filename, data_loader, criterion, print_freq)
def	export_onnx (model, onnx_filename, batch_onnx, per_channel_quantization)
def	calibrate_model (model, model_name, data_loader, num_calib_batch, calibrator, hist_percentile, out_dir)
def	collect_stats (model, data_loader, num_batches)
def	compute_amax (model, **kwargs)
def	finetune_model (model, data_loader)
def	build_sensitivity_profile (model, criterion, data_loader_test)

Variables
def	res = main(sys.argv[1:])

Function Documentation

get_parser()

def classification_flow.get_parser

(

)

Creates an argument parser.

Here is the caller graph for this function:

prepare_model()

def classification_flow.prepare_model	(		model_name,
			data_dir,
			per_channel_quantization,
			batch_size_train,
			batch_size_test,
			batch_size_onnx,
			calibrator,
			pretrained = True,
			ckpt_path = None,
			ckpt_url = None
	)

Prepare the model for the classification flow.
Arguments:
    model_name: name to use when accessing torchvision model dictionary
    data_dir: directory with train and val subdirs prepared "imagenet style"
    per_channel_quantization: iff true use per channel quantization for weights
                               note that this isn't currently supported in ONNX-RT/Pytorch
    batch_size_train: batch size to use when training
    batch_size_test: batch size to use when testing in Pytorch
    batch_size_onnx: batch size to use when testing with ONNX-RT
    calibrator: calibration type to use (max/histogram)

    pretrained: if true a pretrained model will be loaded from torchvision
    ckpt_path: path to load a model checkpoint from, if not pretrained
    ckpt_url: url to download a model checkpoint from, if not pretrained and no path was given
    * at least one of {pretrained, path, url} must be valid

The method returns a the following list:
    [
        Model object,
        data loader for training,
        data loader for Pytorch testing,
        data loader for onnx testing
    ]

Here is the caller graph for this function:

main()

def classification_flow.main

(

cmdline_args

)

Here is the call graph for this function:

evaluate_onnx()

def classification_flow.evaluate_onnx	(		onnx_filename,
			data_loader,
			criterion,
			print_freq
	)

Evaluate accuracy on the given ONNX file using the provided data loader and criterion.
   The method returns the average top-1 accuracy on the given dataset.

Here is the caller graph for this function:

export_onnx()

def classification_flow.export_onnx	(		model,
			onnx_filename,
			batch_onnx,
			per_channel_quantization
	)

Here is the caller graph for this function:

calibrate_model()

def classification_flow.calibrate_model	(		model,
			model_name,
			data_loader,
			num_calib_batch,
			calibrator,
			hist_percentile,
			out_dir
	)

    Feed data to the network and calibrate.
    Arguments:
        model: classification model
        model_name: name to use when creating state files
        data_loader: calibration data set
        num_calib_batch: amount of calibration passes to perform
        calibrator: type of calibration to use (max/histogram)
        hist_percentile: percentiles to be used for historgram calibration
        out_dir: dir to save state files in

Here is the call graph for this function:

Here is the caller graph for this function:

collect_stats()

def classification_flow.collect_stats	(		model,
			data_loader,
			num_batches
	)

Feed data to the network and collect statistics

Here is the caller graph for this function:

compute_amax()

def classification_flow.compute_amax	(		model,
		**	kwargs
	)

Here is the caller graph for this function:

finetune_model()

def classification_flow.finetune_model	(		model,
			data_loader
	)

Finetune the model

Here is the caller graph for this function:

build_sensitivity_profile()

def classification_flow.build_sensitivity_profile	(		model,
			criterion,
			data_loader_test
	)

Here is the caller graph for this function:

Variable Documentation

res

def classification_flow.res = main(sys.argv[1:])