def __init__(self, block, layer_num, num_classes=100):
super(ResNet, self).__init__()
self.conv1 = conv7x7(3, 64, stride=2, padding=3)
self.bn1 = bn_with_initialize(64)
self.relu = P.ReLU()
self.maxpool = MaxPool2d(kernel_size=3, stride=2, pad_mode="same")
self.layer1 = MakeLayer0(block,
layer_num[0],
in_channels=64,
out_channels=256,
stride=1)
self.layer2 = MakeLayer1(block,
layer_num[1],
in_channels=256,
out_channels=512,
stride=2)
self.layer3 = MakeLayer2(block,
layer_num[2],
in_channels=512,
out_channels=1024,
stride=2)
self.layer4 = MakeLayer3(block,
layer_num[3],
in_channels=1024,
out_channels=2048,
stride=2)
self.pool = nn.AvgPool2d(7, 1)
self.fc = fc_with_initialize(512 * block.expansion, num_classes)
self.flatten = Flatten()
def __init__(self, block, num_classes=100):
super(ResNetModelParallel, self).__init__()
self.relu = P.ReLU().shard(((1, dev_num, 1, 1),))
self.maxpool = MaxPool2d(kernel_size=3, stride=2, pad_mode="same")
self.layer1 = MakeLayer0(
block, in_channels=64, out_channels=256, stride=1)
self.pool = M.ReduceMean(keep_dims=True).shard(strategy_no_weight)
self.fc = fc_with_initialize(64 * block.expansion, num_classes)
self.flatten = Flatten()
def __init__(self, num_classes=100):
super(ResNet, self).__init__()
strategy_no_weight = ((dev_num, 1, 1, 1),)
self.conv1 = conv7x7(3, 64, stride=2, padding=0)
self.bn1 = bn_with_initialize(64)
self.relu = ReLU()
self.relu.relu.shard(strategy_no_weight)
self.maxpool = MaxPool2d(kernel_size=3, stride=2, pad_mode="same")
self.reshape = P.Reshape()
self.matmul = P.MatMul().shard(((8, 1), (1, 1)))
self.matmul_weight = Parameter(Tensor(np.ones([200704, num_classes]), dtype=ms.float32), name="weight")
def __init__(self, block, num_classes=100):
super(ResNet, self).__init__()
self.conv1 = conv7x7(3, 64, stride=2)
self.bn1 = bn_with_initialize(64)
self.relu = P.ReLU().set_strategy(strategy_no_weight)
self.maxpool = MaxPool2d(kernel_size=3, stride=2, pad_mode="same")
self.layer1 = MakeLayer0(block,
in_channels=64,
out_channels=256,
stride=1)
self.pool = M.ReduceMean(
keep_dims=True).set_strategy(strategy_no_weight)
self.fc = fc_with_initialize(64 * block.expansion, num_classes)
self.flatten = Flatten()