preProbSquared = x * x
l1loss = torch.sum(0.01 * torch.abs(preProbSquared))
window_size = (int(x.size()[2]), int(x.size()[3]))
x = torch.nn.functional.avg_pool2d(x, kernel_size=window_size)
x = torch.squeeze(x)
probs = self.softmax(x)
logprobs = torch.log(probs)
nll = nllloss(logprobs, labels)
loss = l1loss + nll
return loss
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
c10driver.run(torchWriter, willowOptPatterns, args.outputdir, cifarInIndices,
args.device, args.hw_id)
b = i[0:1].view(3, 1, 4)
h0 = b[0:1, 0:4].view(1, 1, 4)
c0 = b[1:2, 0:4].view(1, 1, 4)
x = self.lstm(a, (h0, c0))
result = torch.cat((x[0], x[1][0], x[1][1]))
l1loss = torch.sum(0.1 * torch.abs(result))
return l1loss
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
c10driver.run(
torchWriter,
popart.Patterns(['PreUniRepl', 'MulArgGradOp', 'AbsGradOp',
'OpToReshape']), args.outputdir, cifarInIndices,
args.device, args.hw_id)
stride=1,
padding=(1, 3),
bias=False)
self.relu = torch.nn.functional.relu
def forward(self, inputs):
"""out = l1loss(relu(conv(in)))"""
image0 = inputs[0]
x = self.conv1(image0)
x = self.relu(x)
x = torch.sum(0.1 * torch.abs(x)) # l1loss
return x
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
c10driver.run(torchWriter, None, args.outputdir, cifarInIndices, args.device,
args.hw_id)
def forward(self, inputs):
"""out = relu(conv(in))"""
image0 = inputs[0]
x = self.conv1(image0)
x = self.relu(x)
return x
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
optimizer=optimizer,
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=samplesPerBatch)
# Passes if torch and popart models match
c10driver.run(torchWriter=torchWriter,
patterns=None,
outputdir=args.outputdir,
cifarInIndices=cifarInIndices,
device=args.device,
device_hw_id=args.hw_id,
mode="infer")
im0 = inputs[0]
x = self.conv1(im0)
x = self.in2(x)
x = self.sin(x)
return x
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
losses=losses,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
c10driver.run(torchWriter,
None,
args.outputdir,
cifarInIndices,
args.device,
args.hw_id,
transformations=["prepareNodesForTraining"],
epochs=4)
# print(x.shape, inputs[2].shape)
loss = nllloss(probs, inputs[2])
return loss
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
# As part of T16818 (model indeterminism) it is useful to
# intercept the output directory here with something like
#args.outputdir="/path/to/logging/dir/where/models/written/"
c10driver.run(torchWriter,
willowOptPatterns,
args.outputdir,
cifarInIndices,
args.device,
args.hw_id,
printAnchorArrays=True)
# seq_length 3
# batch_size 1
# input_size 4
a = i[0:1].view(3, 1, 4)
b = i[0:1].view(3, 1, 4)
h0 = b[0:1, 0:4].view(1, 1, 4)
c0 = b[1:2, 0:4].view(1, 1, 4)
x = self.lstm(a, (h0, c0))
result = x[0]
return result
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
torchWriter = torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
losses=losses,
optimizer=popart.ConstSGD(0.001),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize)
c10driver.run(torchWriter, popart.Patterns(['PreUniRepl']), args.outputdir,
cifarInIndices, args.device, args.hw_id)
# Set arbitrary seed so model weights are initialized to the
# same values each time the test is run
torch.manual_seed(1)
# Test with ConstSGD
anchors_1 = c10driver.run(
torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
losses=losses,
# default loss scaling (1.0f)
optimizer=popart.SGD({"defaultLearningRate": (0.001, True)}),
inputShapeInfo=inputShapeInfo,
dataFlow=dataFlow,
### Torch specific:
module=Module0(),
samplesPerBatch=batchSize),
willowOptPasses,
args.outputdir,
cifarInIndices,
args.device,
args.hw_id)
# Test with ConstSGD
anchors_2 = c10driver.run(
torchwriter.PytorchNetWriter(
inNames=inNames,
outNames=outNames,
losses=losses,
