def forward(self, x):
x = self.bn(self.conv(x))
if self.act == "leaky_relu":
x = F.leaky_relu(x)
elif self.act == "hard_swish":
x = F.hardswish(x)
return x
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
if self.if_act:
if self.act == "relu":
x = F.relu(x)
elif self.act == "hardswish":
x = hardswish(x)
else:
print("The activation function is selected incorrectly.")
exit()
return x
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
if self.act is not None:
if self.act == "relu":
x = F.relu(x)
elif self.act == "relu6":
x = F.relu6(x)
elif self.act == "hard_swish":
x = F.hardswish(x)
else:
raise NotImplementedError(
"The activation function is selected incorrectly.")
return x
def setUp(self):
paddle.disable_static()
self.place = paddle.NPUPlace(0)
self.dtype = np.float32
self.x = np.random.uniform(-6, 10, [8, 15]).astype(self.dtype)
paddle.set_device('cpu')
data = paddle.to_tensor(self.x, stop_gradient=False)
y = F.hardswish(data)
y.sum().backward()
self.out_g = data.grad
self.out_y = y
def forward(self, inputs, label=None):
x = self.conv1(inputs)
for block in self.block_list:
x = block(x)
x = self.last_second_conv(x)
x = self.pool(x)
x = self.last_conv(x)
x = hardswish(x)
x = self.dropout(x)
x = paddle.flatten(x, start_axis=1, stop_axis=-1)
x = self.out(x)
return x
def test_check_output_and_grad_npu(self):
paddle.set_device('npu')
data = paddle.to_tensor(self.x, stop_gradient=False)
y = F.hardswish(data)
y.sum().backward()
self.assertTrue(
np.allclose(self.out_y.numpy(), y.numpy()),
"Output of NPU HardSwish forward has diff at " + str(self.place) +
"\nExpect " + str(self.out_y) + "\n" + "But Got" + str(y) +
" in class " + self.__class__.__name__ + ".")
self.assertTrue(
np.allclose(self.out_g.numpy(), data.grad.numpy()),
"Output of NPU HardSwish backward has diff at " + str(self.place) +
"\nExpect " + str(self.out_g) + "\n" + "But Got" + str(data.grad) +
" in class " + self.__class__.__name__ + ".")
def act_func(self, x):
if self.act == "leaky_relu":
x = F.leaky_relu(x)
elif self.act == "hard_swish":
x = F.hardswish(x)
return x
