def forward(self, inps):
src = inps[0]
weight = inps[1]
try:
bias = inps[2]
except IndexError:
bias = None
if bias is not None:
if bias.shape.ndim == 3:
bias = F.expand_dims(bias, axis=0)
elif bias.shape.ndim == 1:
bias = F.expand_dims(bias, axis=[0, 2, 3])
else:
raise Exception(f"Invalid Conv2d bias's shape {bias.shape}")
if self.param["groups"] != 1:
groups = self.param["groups"]
IC = src.shape.numpy()[1]
OC = weight.shape.numpy()[0]
FH = weight.shape.numpy()[2]
FW = weight.shape.numpy()[3]
target_shape = [groups, int(OC / groups), int(IC / groups), FH, FW]
weight = F.reshape(weight, target_shape)
return F.conv2d(
src,
weight,
bias,
stride=self.param["stride"],
padding=self.param["padding"],
dilation=self.param["dilation"],
groups=self.param["groups"],
)
def forward(self, x):
x1 = self.block_0(x)
x2 = self.block_1(x)
y = x1 + x2
y = F.reshape(y, (-1))
y = y * 3
return y
def test_host_compute_elemwise():
a = mge.tensor([[1, 2], [2, 3], [3, 4]], dtype="float32")
b = mge.tensor([1, 1], dtype="int32")
# check DepType::VALUE is still known
c = b * 2
with pytest.raises(RuntimeError):
d = F.reshape(a, c)
def test_level1_infer_shape_with_unknown():
config_async_level(2)
a = mge.tensor([[1, 2, 2, 3]], dtype="float32")
b = mge.tensor([1, 1])
multi2 = mge.tensor(np.array([[2, 0], [0, 2]]), dtype="float32")
c = F.matmul(b, multi2)
# make DepType::SHAPE unknown
d = F.reshape(a, c)
e = mge.tensor([[1, 2]], dtype="float32")
config_async_level(1)
# test src no shape, throw in level1
with pytest.raises(RuntimeError):
f = F.reshape(d, b)
with pytest.raises(RuntimeError):
g = F.matmul(d, e)
config_async_level(2)
def test_level1_infer_value():
config_async_level(1)
a = mge.tensor([[1, 2], [2, 3], [3, 4]], dtype="float32")
b = mge.tensor([1, 1], dtype="float32")
# make DepType::VALUE unknown
c = b * 2
with pytest.raises(RuntimeError):
d = F.reshape(a, c)
def forward(self, inps):
target_shape = inps[1].numpy()
valid_test = [i for i, k in enumerate(target_shape) if k == -1]
assert (
len(valid_test) <=
1), "Target Shape of ReShape Opr only contains '-1' Once At Most"
if 0 not in target_shape and -1 not in target_shape:
return F.reshape(inps[0], target_shape)
else:
tshape = []
inp_shape = inps[0].shape.numpy()
for i, k in enumerate(target_shape):
if k == 0:
tshape.append(inp_shape[i])
else:
tshape.append(k)
return F.reshape(inps[0], tshape)
def forward(self, input):
A = input.shape[0]
shape = astensor1d((A, A),
self.data,
dtype="int32",
device=input.device)
x = F.reshape(self.data, shape)
o = input + x
return o
def forward(self, x):
x = F.reshape(x, (1, 3, 224, 224))
x = self.extract_features(x)["res5"]
x = F.avg_pool2d(x, 7)
x = F.flatten(x, 1)
x = self.fc(x)
return x
def test_level1_infer_value():
config_async_level(1)
a = mge.tensor([[1, 2], [2, 3], [3, 4]], dtype="float32")
b = mge.tensor([1, 1], dtype="float32")
identity = mge.tensor(np.array([[1, 0], [0, 1]]), dtype="float32")
# make DepType::VALUE unknown
c = F.matmul(b, identity)
with pytest.raises(RuntimeError):
d = F.reshape(a, c)
config_async_level(2)
def test_level1_infer_shape_with_unknown():
config_async_level(2)
a = mge.tensor([[1, 2, 2, 3]], dtype="float32")
b = mge.tensor([1, 1])
c = b * 2
# make DepType::SHAPE unknown
d = F.reshape(a, c)
config_async_level(1)
e = mge.tensor([[1, 2]], dtype="float32")
with pytest.raises(RuntimeError):
f = F.matmul(d, e)
def test_reshape():
x = np.arange(6, dtype="float32")
xx = tensor(x)
y = x.reshape(1, 2, 3)
for shape in [
(1, 2, 3),
(1, -1, 3),
(1, tensor(-1), 3),
np.array([1, -1, 3], dtype="int32"),
tensor([1, -1, 3]),
]:
yy = F.reshape(xx, shape)
np.testing.assert_equal(yy.numpy(), y)
def __init__(self, output_size=5, hidden_size=64, layers=4):
super().__init__()
self.hidden_size = hidden_size
self.base = ConvBase(output_size=hidden_size,
hidden=hidden_size,
channels=1,
max_pool=False,
layers=layers)
self.features = M.Sequential(
kl.Lambda(lambda x: F.reshape(x, (-1, 1, 28, 28))),
self.base,
kl.Lambda(lambda x: kf.mean(x, axis=[2, 3])),
kl.Flatten(),
)
self.classifier = M.Linear(hidden_size, output_size, bias=True)
def test_reshape(is_varnode):
if is_varnode:
network = Network()
else:
network = None
x = np.arange(6, dtype="float32")
xx = make_tensor(x, network)
y = x.reshape(1, 2, 3)
for shape in [
(1, 2, 3),
(1, -1, 3),
(1, make_tensor(-1, network), 3),
np.array([1, -1, 3], dtype="int32"),
make_tensor([1, -1, 3], network),
]:
yy = F.reshape(xx, shape)
np.testing.assert_equal(yy.numpy(), y)
def forward(self, now_LR, pre_h_SD):
"""
now_LR: B,3,H,W
pre_h_SD: B,48,H,W
"""
batch, C, H, W = pre_h_SD.shape
kernels = self.conv(now_LR) # [B, k*k, H, W]
batchwise_ans = []
for idx in range(batch):
kernel = kernels[idx] # [k*k, H, W]
kernel = F.dimshuffle(kernel, (1, 2, 0)) # [H, W , k*k]
kernel = F.reshape(kernel, (H, W, 1, self.K, self.K, 1))
kernel = F.broadcast_to(kernel, (C, H, W, 1, self.K, self.K, 1))
batchwise_ans.append(
F.local_conv2d(
F.add_axis(pre_h_SD[idx], 0), kernel, [1, 1], [1, 1],
[1, 1])) # [1, C, H, W] some bug with padding
similarity_matrix = F.concat(batchwise_ans, axis=0) # [B,C,H,W]
del batchwise_ans
similarity_matrix = F.sigmoid(similarity_matrix)
return F.multiply(pre_h_SD, similarity_matrix)
def forward(self, x):
x = F.reshape(x, self.out_shape)
x = F.reshape(x, self.out_shape1)
x = F.reshape(x, self.out_shape2)
return x
def convert_to_nchw4(var):
var = F.reshape(var, (var.shape[0], var.shape[1] // 4, 4,
var.shape[2], var.shape[3]))
var = F.transpose(var, (0, 1, 3, 4, 2))
return var
def func(x, shp):
return F.reshape(x, shp)
