I’m trying to train PreActResNet-18 with QAT (quantization aware training). Is this the correct way to insert quant
and dequant
stubs into the PreActBlock
and CIFAR_ResNet
definitions? In particular, I am unsure about the lines to do with shortcut
connections in PreActBlock. Thanks!
class PreActBlock(nn.Module):
'''Pre-activation version of the BasicBlock.'''
expansion = 1
def __init__(self, in_planes, planes, stride=1, quant=False):
super(PreActBlock, self).__init__()
self.bn1 = nn.BatchNorm2d(in_planes)
self.conv1 = conv3x3(in_planes, planes, stride)
self.bn2 = nn.BatchNorm2d(planes)
self.conv2 = conv3x3(planes, planes)
self.relu = nn.ReLU(inplace=True)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False)
)
self.Q = quant
if quant:
self.quant = torch.ao.quantization.QuantStub()
self.dequant = torch.ao.quantization.DeQuantStub()
def forward(self, x):
out = self.relu(self.bn1(x))
shortcut = self.shortcut(out)
out = self.conv1(out)
out = self.conv2(self.relu(self.bn2(out)))
if self.Q:
out = self.dequant(out)
shortcut = self.dequant(shortcut)
out += shortcut
if self.Q:
out = self.quant(out)
return out
class CIFAR_ResNet(nn.Module):
def __init__(self, block, num_blocks, num_classes=10, bias=True, in_channels=3, zero_init_residual=False, quant=False):
super(CIFAR_ResNet, self).__init__()
self.Q = quant
self.in_planes = 64
self.conv1 = conv3x3(in_channels,64)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2)
self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2)
self.gap = nn.AvgPool2d(4)
self.linear = nn.Linear(512*block.expansion, num_classes, bias=bias)
if quant:
self.quant = torch.ao.quantization.QuantStub()
self.dequant = torch.ao.quantization.DeQuantStub()
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.in_planes, planes, stride, quant=self.Q))
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x, lin=0, lout=5):
bb = x.size(0)
if self.Q:
x = self.quant(x)
x = self.relu(self.bn1(self.conv1(x)))
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.gap(x)
x = x.view(x.size(0), -1)
x = self.linear(x)
if self.Q:
x = self.dequant(x)
return x