def test_qat_mobile_per_channel_ff():
network = mobilenetV2(num_classes=1000)
img = Tensor(np.ones((1, 3, 224, 224)).astype(np.float32))
network = qat.convert_quant_network(network, bn_fold=True, per_channel=[False, False], symmetric=[True, False])
# should load the checkpoint. mock here
network.init_parameters_data()
qat.export(network, img, file_name="quant.pb")
def test_qat_lenet():
img = Tensor(np.ones((32, 1, 32, 32)).astype(np.float32))
net = LeNet5()
net = qat.convert_quant_network(
net, bn_fold=True, per_channel=[True, False], symmetric=[True, False])
# should load the checkpoint. mock here
net.init_parameters_data()
qat.export(net, img, file_name="quant.pb")
def export_lenet():
context.set_context(mode=context.GRAPH_MODE, device_target=device_target)
cfg = quant_cfg
# define fusion network
network = LeNet5Fusion(cfg.num_classes)
# convert fusion network to quantization aware network
network = quant.convert_quant_network(network,
quant_delay=0,
bn_fold=False,
freeze_bn=10000,
per_channel=[True, False],
symmetric=[True, False])
# export network
inputs = Tensor(np.ones([1, 1, cfg.image_height, cfg.image_width]),
mstype.float32)
quant.export(network,
inputs,
file_name="lenet_quant",
file_format='MINDIR')
from src.config import mnist_cfg as cfg
from src.lenet_fusion import LeNet5 as LeNet5Fusion
parser = argparse.ArgumentParser(description='MindSpore MNIST Example')
parser.add_argument('--device_target', type=str, default="Ascend",
choices=['Ascend', 'GPU'],
help='device where the code will be implemented (default: Ascend)')
parser.add_argument('--data_path', type=str, default="./MNIST_Data",
help='path where the dataset is saved')
parser.add_argument('--ckpt_path', type=str, default="",
help='if mode is test, must provide path where the trained ckpt file')
parser.add_argument('--dataset_sink_mode', type=bool, default=True,
help='dataset_sink_mode is False or True')
args = parser.parse_args()
if __name__ == "__main__":
context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
# define fusion network
network = LeNet5Fusion(cfg.num_classes)
# convert fusion network to quantization aware network
network = quant.convert_quant_network(network, quant_delay=0, bn_fold=False, freeze_bn=10000)
# load quantization aware network checkpoint
param_dict = load_checkpoint(args.ckpt_path)
load_param_into_net(network, param_dict)
# export network
inputs = Tensor(np.ones([1, 1, cfg.image_height, cfg.image_width]), mindspore.float32)
quant.export(network, inputs, file_name="lenet_quant", file_format='AIR')
from src.mobilenetV2 import mobilenetV2
from src.config import config_ascend_quant
parser = argparse.ArgumentParser(description='Image classification')
parser.add_argument('--checkpoint_path', type=str, default=None, help='Checkpoint file path')
parser.add_argument('--device_target', type=str, default=None, help='Run device target')
args_opt = parser.parse_args()
if __name__ == '__main__':
cfg = None
if args_opt.device_target == "Ascend":
cfg = config_ascend_quant
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", save_graphs=False)
else:
raise ValueError("Unsupported device target: {}.".format(args_opt.device_target))
# define fusion network
network = mobilenetV2(num_classes=cfg.num_classes)
# convert fusion network to quantization aware network
network = quant.convert_quant_network(network, bn_fold=True, per_channel=[True, False], symmetric=[True, False])
# load checkpoint
param_dict = load_checkpoint(args_opt.checkpoint_path)
load_param_into_net(network, param_dict)
# export network
print("============== Starting export ==============")
inputs = Tensor(np.ones([1, 3, cfg.image_height, cfg.image_width]), mindspore.float32)
quant.export(network, inputs, file_name="mobilenet_quant", file_format='MINDIR')
print("============== End export ==============")