def test_invalid_multiconv_options():
mc = poptorch.MultiConv()
with pytest.raises(AssertionError, match="Invalid partials types"):
mc.partialsTypes("half")
with pytest.raises(AssertionError, match="Invalid plan type"):
mc.planType("parallel")
def test_multiconv_options_broadcast():
multiconv = (
poptorch.MultiConv().availableMemoryProportions(0.8).partialsTypes(
poptorch.MultiConvPartialsType.Float).planType(
poptorch.MultiConvPlanType.Parallel).perConvReservedTiles(
100).cycleBackOff(0.3))
multiconv_harness(multiconv)
def test_multiconv_options_per_conv():
partials_types = [
poptorch.MultiConvPartialsType.Float,
poptorch.MultiConvPartialsType.Float
]
multiconv = (poptorch.MultiConv().availableMemoryProportions(
(0.8, 0.7)).partialsTypes(partials_types).planType(
poptorch.MultiConvPlanType.Parallel).perConvReservedTiles(
120).cycleBackOff(0.4))
multiconv_harness(multiconv)
def forward(self, x):
with poptorch.MultiConv():
y = self.layer1A(x)
z = self.layer1B(x)
x = y + z
x = self.layer2(x)
x = x.view(-1, 320)
x = self.layer3_act(self.layer3(x))
x = self.layer4(x)
x = self.softmax(x)
return x
def forward(self, x, y):
with poptorch.MultiConv():
return (self.convA(x), self.convB(y))
def forward(self, x):
with poptorch.MultiConv():
return torch.pow(x, 2)
def forward(self, x):
with poptorch.MultiConv():
with poptorch.MultiConv():
return self.conv(x)
def forwardWithMultiConv(*args, **kwargs):
with poptorch.MultiConv():
return forward_impl(*args, **kwargs)