• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2019 Huawei Technologies Co., Ltd
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14
15import numpy as np
16
17import mindspore as ms
18import mindspore.nn as nn
19from mindspore import Tensor, Parameter, ParameterTuple
20from mindspore import context
21from mindspore.common.api import _cell_graph_executor
22from mindspore.ops import composite as C
23from mindspore.ops import operations as P
24
25
26grad_by_list = C.GradOperation(get_by_list=True)
27
28
29class NetWithLoss(nn.Cell):
30    def __init__(self, network, strategy3):
31        super(NetWithLoss, self).__init__()
32        self.loss = P.SoftmaxCrossEntropyWithLogits().shard(strategy3)
33        self.network = network
34
35    def construct(self, x, b):
36        predict = self.network(x)
37        return self.loss(predict, b)[0]
38
39
40class OneStepCell(nn.Cell):
41    def __init__(self, network):
42        super(OneStepCell, self).__init__(auto_prefix=False)
43        self.network = network
44        self.weights = ParameterTuple(network.network.trainable_params())
45
46    def construct(self, data, label):
47        weights = self.weights
48        grads = grad_by_list(self.network, weights)(data, label)
49        return grads
50
51
52def test_one_weight_parameter():
53    class Net(nn.Cell):
54        def __init__(self, strategy1, weight):
55            super().__init__()
56            self.weight = Parameter(weight, "w1", requires_grad=True)
57            self.matmul = P.MatMul().shard(strategy1)
58
59        def construct(self, x):
60            out = self.matmul(x, self.weight)
61            return out
62
63    context.set_auto_parallel_context(device_num=8, global_rank=0)
64    strategy1 = ((4, 1), (1, 2))
65    strategy3 = ((8, 1), (8, 1))
66
67    x = Tensor(np.ones([64, 32]), dtype=ms.float32)
68    weight = Tensor(np.ones([32, 64]), dtype=ms.float32)
69    b = Tensor(np.ones([64, 64]), dtype=ms.float32)
70
71    net = Net(strategy1, weight)
72
73    net_with_loss = NetWithLoss(net, strategy3)
74
75    train_net = OneStepCell(net_with_loss)
76    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
77    train_net.set_auto_parallel()
78
79    train_net.set_train()
80    _cell_graph_executor.compile(train_net, x, b)
81