"""
-------------------------------------------------------------------------
This file is part of the MindStudio project.
Copyright (c) 2025 Huawei Technologies Co.,Ltd.
MindStudio is licensed under Mulan PSL v2.
You can use this software according to the terms and conditions of the Mulan PSL v2.
You may obtain a copy of Mulan PSL v2 at:
http://license.coscl.org.cn/MulanPSL2
THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
See the Mulan PSL v2 for more details.
-------------------------------------------------------------------------
"""
import mindspore.nn as nn
from mindspore.ops import operations as P
from mindformers.modules.layers import Linear
from mindformers.models.modeling_utils import PreTrainedModel
from mindformers.models.configuration_utils import PretrainedConfig
def conv_bn_relu(in_channel, out_channel, kernel_size, stride, depth_wise, activation='relu6'):
output = [nn.Conv2d(in_channel, out_channel, kernel_size, stride, pad_mode="same",
group=1 if not depth_wise else in_channel), nn.BatchNorm2d(out_channel)]
if activation:
output.append(nn.get_activation(activation))
return nn.SequentialCell(output)
class TestNetMindSpore(nn.Cell):
"""
MobileNet V1 backbone
"""
def __init__(self, class_num=10, features_only=False):
super(TestNetMindSpore, self).__init__()
self.features_only = features_only
cnn1 = [
conv_bn_relu(3, 32, 3, 2, False),
]
cnn2 = [
conv_bn_relu(32, 32, 3, 2, False),
]
self.network1 = nn.SequentialCell(cnn1)
self.network2 = nn.SequentialCell(cnn2)
self.fc = nn.Dense(32, class_num)
def construct(self, x):
output = x
output = self.network1(output)
output = self.network2(output)
output = P.ReduceMean()(output, (2, 3))
output = self.fc(output)
return output
class TestNetMindSpore2(nn.Cell):
"""
MobileNet V1 backbone
"""
def __init__(self, class_num=10, features_only=False):
super(TestNetMindSpore2, self).__init__()
self.features_only = features_only
cnn = [
conv_bn_relu(3, 32, 3, 2, False),
]
self.backbone = nn.SequentialCell(cnn)
self.fc = nn.Dense(32, class_num)
def construct(self, x):
output = x
output = self.backbone(output)
output = P.ReduceMean()(output, (2, 3))
output = self.fc(output)
return output
class TestNetMindSpore3(nn.Cell):
"""
MobileNet V1 backbone
"""
def __init__(self, class_num=10, features_only=False):
super().__init__()
self.features_only = features_only
cnn = [
conv_bn_relu(3, 32, 3, 2, False),
]
self.backbone = nn.SequentialCell(cnn)
self.fc = nn.Dense(32, class_num)
def construct(self, x, y):
output = x
output = self.backbone(output)
output = P.ReduceMean()(output, (2, 3))
output = self.fc(output)
return output
class TestNetMindSpore4(nn.Cell):
"""
TestNet
"""
def __init__(self, class_num=10):
super().__init__()
self.backbone = nn.SequentialCell([conv_bn_relu(3, 64, 3, 2, False),
conv_bn_relu(64, 32, 3, 1, False)])
self.fc = nn.Dense(32, class_num)
def forward(self, x):
x = self.backbone(x)
x = P.ReduceMean(keep_dims=True)(x)
x = x.Flatten()
x = self.fc(x)
return x
def get_model():
return TestNetMindSpore(class_num=10)
class SampleModel(nn.Cell):
def __init__(self, num_class=10, num_channel=3):
super(SampleModel, self).__init__()
self.conv1 = nn.Conv2d(num_channel, 16, 3, pad_mode='valid')
self.conv2 = nn.Conv2d(16, 16, 3, pad_mode='valid')
self.fc1 = nn.Dense(16 * 3 * 3, 10)
self.fc2 = nn.Dense(10, num_class)
self.relu = nn.ReLU()
def construct(self, x):
x = self.conv1(x)
x = self.relu(x)
x = self.conv2(x)
y = self.relu(x)
return y
class LrdSampleNetwork(nn.Cell):
def __init__(self):
super().__init__()
self.embedding = nn.SequentialCell([
nn.Embedding(16, 32),
nn.Dense(32, 64),
nn.ReLU(),
])
self.feature = nn.SequentialCell([
nn.Conv2d(64, 128, 3, pad_mode="pad", padding=1, has_bias=True),
nn.ReLU(),
nn.Conv2d(128, 64, 1, has_bias=True),
nn.ReLU(),
])
self.pool = nn.AvgPool2d((7, 7), stride=2)
self.flatten = nn.Flatten()
self.inner = nn.Dense(64 * 5 * 5, 512)
self.classifier = nn.SequentialCell([
nn.Dense(512, 256),
nn.Dense(256, 10),
])
def construct(self, inputs):
shortcut = self.embedding(inputs)
shortcut = shortcut.transpose([0, 3, 1, 2])
next_node = self.feature(shortcut)
next_node = next_node + shortcut
next_node = self.pool(next_node)
next_node = self.flatten(next_node)
next_node = self.inner(next_node)
next_node = self.classifier(next_node)
return next_node
class MsTeacherModel(nn.Cell):
def __init__(self):
super().__init__()
self.teacher_fc = nn.Dense(1, 1)
def construct(self, inputs):
logits = self.teacher_fc(inputs)
loss = logits / 2
return loss
class MsStudentModel(nn.Cell):
def __init__(self):
super().__init__()
self.student_fc = nn.Dense(1, 1)
def construct(self, inputs):
logits = self.student_fc(inputs)
loss = logits / 2
return loss
class MsSparseModel(PreTrainedModel):
def __init__(self, config: PretrainedConfig):
super().__init__(config)
self.q = Linear(256, 256)
self.k = Linear(256, 256)
self.v = Linear(256, 256)
def construct(self, x):
x1 = self.q(x)
x2 = self.k(x)
x3 = self.v(x)
return x1 + x2 + x3
