criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(),
lr=args.lr,
momentum=0.9,
weight_decay=5e-4)
# loading wm examples
if args.wmtrain:
print("WM acc:")
test(net, criterion, logfile, wmloader, device)
# start training
for epoch in range(start_epoch, start_epoch + args.max_epochs):
# adjust learning rate
adjust_learning_rate(args.lr, optimizer, epoch, args.lradj, args.ratio)
train(epoch, net, criterion, optimizer, logfile, trainloader, device,
wmloader)
print("Test acc:")
acc = test(net, criterion, logfile, testloader, device)
if args.wmtrain:
print("WM acc:")
test(net, criterion, logfile, wmloader, device)
print('Saving..')
state = {
'net': net.module if device is 'cuda' else net,
'acc': acc,
torch.cuda.device_count()))
parent = torch.nn.DataParallel(parent,
device_ids=range(
torch.cuda.device_count()))
cudnn.benchmark = True
test_criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(),
lr=lr,
momentum=0.9,
weight_decay=5e-4)
# start training
for epoch in range(start_epoch, start_epoch + max_epochs):
# adjust learning rate
adjust_learning_rate(lr, optimizer, epoch, lradj)
train_steal(epoch,
net,
parent,
optimizer,
logfile,
testloader if use_test else trainloader,
device,
grad_query=grad_query)
print("Test acc:")
acc = test(net, test_criterion, logfile, testloader, device)
print('Saving..')
state = {