2020-03-27 14:49:12 +08:00
|
|
|
# Copyright 2019 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
|
2020-05-18 16:42:35 +08:00
|
|
|
|
|
|
|
import mindspore as ms
|
2020-03-27 14:49:12 +08:00
|
|
|
import mindspore.nn as nn
|
|
|
|
from mindspore import Tensor, Parameter
|
2020-05-18 16:42:35 +08:00
|
|
|
from mindspore import context
|
2021-08-27 10:33:35 +08:00
|
|
|
from mindspore.common.api import _CellGraphExecutor
|
2020-03-27 14:49:12 +08:00
|
|
|
from mindspore.nn import TrainOneStepCell
|
2020-05-18 16:42:35 +08:00
|
|
|
from mindspore.nn.optim import AdamWeightDecay
|
|
|
|
from mindspore.ops import operations as P
|
2020-03-27 14:49:12 +08:00
|
|
|
|
|
|
|
|
|
|
|
class NetWithLoss(nn.Cell):
|
|
|
|
def __init__(self, network, strategy3):
|
|
|
|
super(NetWithLoss, self).__init__()
|
2020-09-10 15:30:19 +08:00
|
|
|
self.loss = P.SoftmaxCrossEntropyWithLogits().shard(strategy3)
|
2020-03-27 14:49:12 +08:00
|
|
|
self.network = network
|
|
|
|
|
|
|
|
def construct(self, x, b):
|
|
|
|
predict = self.network(x)
|
|
|
|
return self.loss(predict, b)[0]
|
|
|
|
|
|
|
|
|
2020-05-22 09:13:09 +08:00
|
|
|
def compile_net(net, x, b):
|
2020-05-07 10:40:59 +08:00
|
|
|
net.set_auto_parallel()
|
2021-08-27 10:33:35 +08:00
|
|
|
_CellGraphExecutor().compile(net, x, b)
|
2020-05-07 10:40:59 +08:00
|
|
|
|
|
|
|
|
2020-03-27 14:49:12 +08:00
|
|
|
def test_optimizer_clone_weight():
|
|
|
|
class Net(nn.Cell):
|
|
|
|
def __init__(self, strategy1, strategy2, weight):
|
|
|
|
super().__init__()
|
|
|
|
self.weight = Parameter(weight, "w1")
|
2020-09-10 15:30:19 +08:00
|
|
|
self.matmul = P.MatMul(transpose_a=False, transpose_b=True).shard(strategy1)
|
|
|
|
self.relu = P.ReLU().shard(strategy2)
|
2020-03-27 14:49:12 +08:00
|
|
|
|
|
|
|
def construct(self, x):
|
|
|
|
out = self.matmul(x, self.weight)
|
|
|
|
out = self.relu(out)
|
|
|
|
return out
|
|
|
|
|
|
|
|
context.set_auto_parallel_context(device_num=4, global_rank=0)
|
2020-08-20 09:36:06 +08:00
|
|
|
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
|
2020-05-18 10:31:46 +08:00
|
|
|
|
2020-03-27 14:49:12 +08:00
|
|
|
strategy1 = ((2, 1), (2, 1))
|
2020-05-18 10:31:46 +08:00
|
|
|
strategy2 = ((4, 1),)
|
2020-03-27 14:49:12 +08:00
|
|
|
strategy3 = ((4, 1), (4, 1))
|
|
|
|
|
|
|
|
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
|
|
|
|
weight = Tensor(np.ones([64, 32]), dtype=ms.float32)
|
|
|
|
b = Tensor(np.ones([64, 64]), dtype=ms.float32)
|
|
|
|
|
|
|
|
net = Net(strategy1, strategy2, weight)
|
|
|
|
|
|
|
|
optimizer = AdamWeightDecay(net.trainable_params())
|
|
|
|
|
|
|
|
net_with_loss = NetWithLoss(net, strategy3)
|
|
|
|
|
|
|
|
train_net = TrainOneStepCell(net_with_loss, optimizer)
|
|
|
|
|
2020-05-22 09:13:09 +08:00
|
|
|
compile_net(train_net, x, b)
|
2020-03-27 14:49:12 +08:00
|
|
|
|
|
|
|
|
|
|
|
def test_optimizer_clone_weight2():
|
|
|
|
class Net(nn.Cell):
|
|
|
|
def __init__(self, strategy1, strategy2, weight):
|
|
|
|
super().__init__()
|
|
|
|
self.weight = Parameter(weight, "w1")
|
2020-09-10 15:30:19 +08:00
|
|
|
self.matmul = P.MatMul(transpose_a=False, transpose_b=True).shard(strategy1)
|
|
|
|
self.relu = P.ReLU().shard(strategy2)
|
2020-03-27 14:49:12 +08:00
|
|
|
|
|
|
|
def construct(self, x):
|
|
|
|
out = self.matmul(x, self.weight)
|
|
|
|
out = self.relu(out)
|
|
|
|
return out
|
|
|
|
|
|
|
|
context.set_auto_parallel_context(device_num=4, global_rank=0)
|
2020-08-20 09:36:06 +08:00
|
|
|
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
|
2020-05-18 10:31:46 +08:00
|
|
|
|
2020-03-27 14:49:12 +08:00
|
|
|
strategy1 = ((2, 1), (2, 1))
|
2020-05-18 10:31:46 +08:00
|
|
|
strategy2 = ((4, 1),)
|
2020-03-27 14:49:12 +08:00
|
|
|
strategy3 = ((4, 1), (4, 1))
|
|
|
|
|
|
|
|
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
|
|
|
|
weight = Tensor(np.ones([64, 32]), dtype=ms.float32)
|
|
|
|
b = Tensor(np.ones([64, 64]), dtype=ms.float32)
|
|
|
|
|
|
|
|
net = Net(strategy1, strategy2, weight)
|
|
|
|
|
|
|
|
optimizer = AdamWeightDecay(net.trainable_params())
|
|
|
|
|
|
|
|
net_with_loss = NetWithLoss(net, strategy3)
|
|
|
|
|
|
|
|
train_net = TrainOneStepCell(net_with_loss, optimizer)
|
|
|
|
|
2020-05-22 09:13:09 +08:00
|
|
|
compile_net(train_net, x, b)
|