!42738 add the st of intergration of dynamic and static memory
Merge pull request !42738 from limingqi107/bug_fix3
This commit is contained in:
Gitee
commit
33b94acde2
|
|
@ -179,3 +179,33 @@ def test_fused_cast_adam_weight_decay():
|
|||
res = train_network(data, label)
|
||||
loss.append(res.asnumpy())
|
||||
assert np.all(loss[-1] < 0.1)
|
||||
|
||||
|
||||
@pytest.mark.level1
|
||||
@pytest.mark.platform_arm_ascend_training
|
||||
@pytest.mark.platform_x86_ascend_training
|
||||
@pytest.mark.platform_x86_gpu_training
|
||||
@pytest.mark.env_onecard
|
||||
def test_fused_cast_adam_weight_decay_with_memory_optimize():
|
||||
'''
|
||||
Feature: Integration of dynamic and static memory in the heterogeneous scene
|
||||
Description: Test FusedCastAdamWeightDecay
|
||||
Expectation: Run lenet success
|
||||
'''
|
||||
context.set_context(mode=context.GRAPH_MODE, memory_optimize_level="O1")
|
||||
data = Tensor(np.ones([32, 3, 32, 32]).astype(np.float32) * 0.01)
|
||||
label = Tensor(np.ones([32]).astype(np.int32))
|
||||
net = LeNet()
|
||||
net.batch_size = 32
|
||||
learning_rate = 0.01
|
||||
optimizer = FusedAdamWeightDecayWithGlobalNorm(filter(lambda x: x.requires_grad, net.get_parameters()),
|
||||
learning_rate)
|
||||
criterion = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
|
||||
net_with_criterion = WithLossCell(net, criterion)
|
||||
train_network = TrainOneStepCell(net_with_criterion, optimizer)
|
||||
train_network.set_train()
|
||||
loss = []
|
||||
for _ in range(10):
|
||||
res = train_network(data, label)
|
||||
loss.append(res.asnumpy())
|
||||
assert np.all(loss[-1] < 0.1)
|
||||
|
|
|
|||
|
|
@ -0,0 +1,142 @@
|
|||
# Copyright 2022 Huawei Technologies Co., Ltd
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
import numpy as np
|
||||
import pytest
|
||||
from mindspore import context, nn, Tensor
|
||||
from mindspore.ops import operations as P
|
||||
from mindspore.ops import composite as C
|
||||
from mindspore.common.parameter import Parameter
|
||||
import mindspore.common.dtype as mstype
|
||||
|
||||
|
||||
grad_all = C.GradOperation(get_all=True)
|
||||
|
||||
|
||||
@pytest.mark.level0
|
||||
@pytest.mark.platform_arm_ascend_training
|
||||
@pytest.mark.platform_x86_gpu_training
|
||||
@pytest.mark.env_onecard
|
||||
def test_while_grad_with_memory_optimize():
|
||||
"""
|
||||
Feature: Integration of dynamic and static memory.
|
||||
Description: Test the control flow scene.
|
||||
Expectation: The result meet expectation.
|
||||
"""
|
||||
class MyWhileNet(nn.Cell):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.max = P.ReduceMax()
|
||||
|
||||
def construct(self, idx, end, x):
|
||||
while idx < end:
|
||||
part = x[idx, :, :]
|
||||
max_num = self.max(part)
|
||||
x[idx, :, 0:2] = max_num
|
||||
idx = idx + 1
|
||||
return x
|
||||
|
||||
class GradNet(nn.Cell):
|
||||
def __init__(self, net):
|
||||
super(GradNet, self).__init__()
|
||||
self.net = net
|
||||
|
||||
def construct(self, *inputs):
|
||||
return grad_all(self.net)(*inputs)
|
||||
|
||||
idx = Tensor(np.array(0), dtype=mstype.int32)
|
||||
end = Tensor(np.array(2), dtype=mstype.int32)
|
||||
input_x = np.array([[[4, 0], [0, 0]],
|
||||
[[0, 4], [0, 0]]]).astype(np.float32)
|
||||
x = Tensor(input_x, dtype=mstype.float32)
|
||||
# memory optimize mode
|
||||
context.set_context(mode=context.GRAPH_MODE, memory_optimize_level="O1")
|
||||
while_net = MyWhileNet()
|
||||
net = GradNet(while_net)
|
||||
graph_output = net(idx, end, x)
|
||||
|
||||
expect_zero = np.array([0], dtype=np.float32)
|
||||
expect_two = input_x
|
||||
assert np.allclose(graph_output[0].asnumpy(), expect_zero, 0.0001, 0.0001)
|
||||
assert np.allclose(graph_output[1].asnumpy(), expect_zero, 0.0001, 0.0001)
|
||||
assert np.allclose(graph_output[2].asnumpy(), expect_two, 0.0001, 0.0001)
|
||||
|
||||
|
||||
class SparseApplyFtrlNet(nn.Cell):
|
||||
def __init__(self, var, accum, linear, lr=0.001, l1=0.0, l2=0.0, lr_power=-0.5):
|
||||
super(SparseApplyFtrlNet, self).__init__()
|
||||
self.sparse_apply_ftrl = P.SparseApplyFtrl(lr=lr, l1=l1, l2=l2, lr_power=lr_power)
|
||||
self.var = Parameter(var, name="var")
|
||||
self.accum = Parameter(accum, name="accum")
|
||||
self.linear = Parameter(linear, name="linear")
|
||||
|
||||
def construct(self, grad, indices):
|
||||
out = self.sparse_apply_ftrl(self.var, self.accum, self.linear, grad, indices)
|
||||
return out
|
||||
|
||||
|
||||
@pytest.mark.level0
|
||||
@pytest.mark.platform_x86_cpu
|
||||
@pytest.mark.env_onecard
|
||||
def test_sparse_apply_ftrl_with_memory_optimize():
|
||||
"""
|
||||
Feature: Integration of dynamic and static memory.
|
||||
Description: Test the scene of output ref node.
|
||||
Expectation: The result meet expectation.
|
||||
"""
|
||||
context.set_context(mode=context.GRAPH_MODE, memory_optimize_level="O1")
|
||||
|
||||
grad_np = np.ones([3, 3, 3])
|
||||
indice_np = [0, 1, 2]
|
||||
var_np = np.ones([3, 3, 3])
|
||||
accum_np = np.ones([3, 3, 3])
|
||||
linear_np = np.ones([3, 3, 3])
|
||||
|
||||
# test1: var/accum/linear/gradient are float32 and indices is int32.
|
||||
gradient = Tensor(grad_np, dtype=mstype.float32)
|
||||
indices = Tensor(indice_np, dtype=mstype.int32)
|
||||
var = Tensor(var_np, dtype=mstype.float32)
|
||||
accum = Tensor(accum_np, dtype=mstype.float32)
|
||||
linear = Tensor(linear_np, dtype=mstype.float32)
|
||||
sparse_apply_ftrl = SparseApplyFtrlNet(var, accum, linear)
|
||||
out = sparse_apply_ftrl(gradient, indices)
|
||||
expect_var = np.array([[[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479]],
|
||||
[[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479]],
|
||||
[[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479],
|
||||
[0.291479, 0.291479, 0.291479]]]).astype(np.float32)
|
||||
assert np.all(out[0].asnumpy() == expect_var)
|
||||
|
||||
# test2: var/accum/linear/gradient are float16 and indices is int32.
|
||||
gradient = Tensor(grad_np, dtype=mstype.float16)
|
||||
indices = Tensor(indice_np, dtype=mstype.int32)
|
||||
var = Tensor(var_np, dtype=mstype.float16)
|
||||
accum = Tensor(accum_np, dtype=mstype.float16)
|
||||
linear = Tensor(linear_np, dtype=mstype.float16)
|
||||
sparse_apply_ftrl = SparseApplyFtrlNet(var, accum, linear)
|
||||
out = sparse_apply_ftrl(gradient, indices)
|
||||
expect_var = np.array([[[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915]],
|
||||
[[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915]],
|
||||
[[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915],
|
||||
[0.2915, 0.2915, 0.2915]]]).astype(np.float16)
|
||||
assert np.all(out[0].asnumpy() == expect_var)
|
||||
Loading…
Reference in New Issue