1# Copyright 2020 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 16import mindspore as ms 17import mindspore.nn as nn 18from mindspore import Tensor 19from mindspore import context 20from mindspore.common.api import _cell_graph_executor 21from mindspore.ops import composite as C 22from mindspore.ops import operations as P 23from tests.ut.python.ops.test_math_ops import VirtualLoss 24 25 26grad_all = C.GradOperation(get_all=True) 27 28 29class NetWithLoss(nn.Cell): 30 def __init__(self, network): 31 super(NetWithLoss, self).__init__() 32 self.loss = VirtualLoss() 33 self.network = network 34 35 def construct(self, x, y): 36 predict = self.network(x, y) 37 return self.loss(predict) 38 39 40class GradWrap(nn.Cell): 41 def __init__(self, network): 42 super(GradWrap, self).__init__() 43 self.network = network 44 45 def construct(self, x, y): 46 return grad_all(self.network)(x, y) 47 48 49class Net(nn.Cell): 50 def __init__(self, strategy1=None, strategy2=None): 51 super().__init__() 52 self.dropout = P.Dropout(keep_prob=0.6).shard(strategy1) 53 self.matmul = P.MatMul().shard(strategy2) 54 55 def construct(self, x, y): 56 out = self.matmul(x, y) 57 out, _ = self.dropout(out) 58 return out 59 60 61def test_dropout_semi_auto(): 62 context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode="semi_auto_parallel") 63 net = GradWrap(NetWithLoss(Net())) 64 net.set_auto_parallel() 65 66 x = Tensor(np.ones([64, 32]), dtype=ms.float32) 67 y = Tensor(np.ones([32, 128]), dtype=ms.float32) 68 net.set_train() 69 _cell_graph_executor.compile(net, x, y) 70 71 72def test_dropout_semi_auto2(): 73 context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode="semi_auto_parallel") 74 strategy1 = ((8, 1),) 75 strategy2 = ((4, 2), (2, 1)) 76 net = GradWrap(NetWithLoss(Net(strategy1, strategy2))) 77 net.set_auto_parallel() 78 79 x = Tensor(np.ones([64, 32]), dtype=ms.float32) 80 y = Tensor(np.ones([32, 128]), dtype=ms.float32) 81 net.set_train() 82 _cell_graph_executor.compile(net, x, y) 83 84 85def test_dropout_semi_auto3(): 86 context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode="semi_auto_parallel") 87 strategy1 = ((2, 4),) 88 strategy2 = ((4, 2), (2, 1)) 89 net = GradWrap(NetWithLoss(Net(strategy1, strategy2))) 90 net.set_auto_parallel() 91 92 x = Tensor(np.ones([64, 32]), dtype=ms.float32) 93 y = Tensor(np.ones([32, 128]), dtype=ms.float32) 94 net.set_train() 95 _cell_graph_executor.compile(net, x, y) 96 97 98def test_dropout_auto(): 99 context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode="auto_parallel") 100 net = GradWrap(NetWithLoss(Net())) 101 net.set_auto_parallel() 102 103 x = Tensor(np.ones([64, 32]), dtype=ms.float32) 104 y = Tensor(np.ones([32, 128]), dtype=ms.float32) 105 net.set_train() 106 _cell_graph_executor.compile(net, x, y) 107