# 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 import mindspore as ms import mindspore.nn as nn from mindspore import Tensor, Parameter from mindspore import context from mindspore.common import dtype as mstype from mindspore.common.api import _executor from mindspore.nn.cell import Cell from mindspore.nn.optim.momentum import Momentum from mindspore.ops import composite as C from mindspore.ops import functional as F from mindspore.ops import operations as P from mindspore.train import Model, ParallelMode from tests.dataset_mock import MindData from tests.ut.python.ops.test_math_ops import VirtualLoss device_num = 16 device_id = 2 class StrategyModel(): onehot_strategy = ((1, device_num), (), ()) twod_strategy = ((1, device_num),) twod_strategy_m = ((device_num, 1),) scalar_twod_strategy = ((), (1, device_num)) twod_scalar_strategy = ((1, device_num), ()) scalar_strategy = ((),) oned_strategy = ((1,),) scalar_scalar_strategy = ((), ()) twod_twod_strategy = ((1, device_num), (1, device_num)) twod_twodbc_strategy = ((1, device_num), (1, 1)) twodbc_twod_strategy = ((1, 1), (device_num, 1)) class StrategyBatch(): onehot_strategy = ((device_num, 1), (), ()) twod_strategy = ((1, device_num),) twod_strategy_m = ((device_num, 1),) scalar_twod_strategy = ((), (1, device_num)) twod_scalar_strategy = ((1, device_num), ()) scalar_strategy = ((),) oned_strategy = ((1,),) scalar_scalar_strategy = ((), ()) twod_twod_strategy = ((1, device_num), (1, device_num)) twod_twodbc_strategy = ((1, device_num), (1, 1)) twodbc_twod_strategy = ((1, 1), (device_num, 1)) class Args(): a = 1 b = 2 c = 3 d = 4 e = 5 num_classes = 512 emb_size = 512 class SemiAutoOneHotNet(Cell): def __init__(self, args, strategy): super(SemiAutoOneHotNet, self).__init__() self.a = args.a self.b = args.b self.c = args.c self.d = args.d self.e = args.e self.cast = P.Cast() self.cast.set_strategy(strategy=strategy.twod_strategy) self.cast1 = P.Cast() self.cast1.set_strategy(strategy=strategy.twod_strategy) self.cast2 = P.Cast() self.cast2.set_strategy(strategy=strategy.twod_strategy) self.cast3 = P.Cast() self.cast3.set_strategy(strategy=strategy.scalar_strategy) self.cast4 = P.Cast() self.cast4.set_strategy(strategy=strategy.scalar_strategy) self.a_const = Tensor(self.a, dtype=mstype.float32) self.b_const = Tensor(self.b, dtype=mstype.float32) self.c_const = Tensor(self.c, dtype=mstype.float32) self.d_const = Tensor(self.d, dtype=mstype.float32) self.e_const = Tensor(self.e, dtype=mstype.float32) self.m_const_zero = Tensor(0, dtype=mstype.float32) self.a_const_one = Tensor(1, dtype=mstype.float32) self.onehot = P.OneHot() self.onehot.set_strategy(strategy=strategy.onehot_strategy) self.exp = P.Exp() self.exp.set_strategy(strategy=strategy.twod_strategy) self.exp2 = P.Exp() self.exp2.set_strategy(strategy=strategy.twod_strategy) self.exp3 = P.Exp() self.exp3.set_strategy(strategy=strategy.twod_strategy) self.mul_const = P.Mul() self.mul_const.set_strategy(strategy=strategy.scalar_twod_strategy) self.mul_const2 = P.TensorAdd() self.mul_const2.set_strategy(strategy=strategy.scalar_twod_strategy) self.mul_const3 = P.Sub() self.mul_const3.set_strategy(strategy=strategy.twod_scalar_strategy) self.mul_const4 = P.Sub() self.mul_const4.set_strategy(strategy=strategy.scalar_twod_strategy) self.mul_const5 = P.Mul() self.mul_const5.set_strategy(strategy=strategy.twod_scalar_strategy) self.mul = P.Mul() self.mul.set_strategy(strategy=strategy.twod_twod_strategy) self.mul2 = P.Mul() self.mul2.set_strategy(strategy=strategy.twod_twod_strategy) self.mul3 = P.TensorAdd() self.mul3.set_strategy(strategy=strategy.twod_twod_strategy) self.mul4 = P.Sub() self.mul4.set_strategy(strategy=strategy.twod_twodbc_strategy) self.mul5 = P.RealDiv() self.mul5.set_strategy(strategy=strategy.twod_twodbc_strategy) self.mul6 = P.Mul() self.mul6.set_strategy(strategy=strategy.twod_twod_strategy) self.mul7 = P.Mul() self.mul7.set_strategy(strategy=strategy.twod_scalar_strategy) self.mul8 = P.RealDiv() self.mul8.set_strategy(strategy=strategy.scalar_scalar_strategy) self.mul9 = P.TensorAdd() self.mul9.set_strategy(strategy=strategy.twod_scalar_strategy) self.reduce_max = P.ReduceMax(keep_dims=True) self.reduce_max.set_strategy(strategy=strategy.twod_strategy) self.reduce_sum = P.ReduceSum(keep_dims=False) self.reduce_sum.set_strategy(strategy=strategy.twod_strategy) self.reduce_sum_2 = P.ReduceSum(keep_dims=False) self.reduce_sum_2.set_strategy(strategy=strategy.twod_strategy) self.reduce_sum_3 = P.ReduceSum(keep_dims=False) self.reduce_sum_3.set_strategy(strategy=strategy.oned_strategy) self.reshape = P.Reshape() self.log = P.Log() self.log.set_strategy(strategy=strategy.twod_strategy) self.on_value = Tensor(1.0, mstype.float32) self.off_value = Tensor(0.0, mstype.float32) self.normalize = P.L2Normalize(axis=1) self.normalize.set_strategy(strategy=strategy.twod_strategy_m) self.normalize2 = P.L2Normalize(axis=1) self.normalize2.set_strategy(strategy=strategy.twod_strategy_m) self.fc = P.MatMul(transpose_b=True) self.fc.set_strategy(strategy=strategy.twodbc_twod_strategy) weight_shape = [args.num_classes, args.emb_size] weight_np = np.zeros(weight_shape, np.float32) self.weight = Parameter(Tensor(weight_np), name='model_parallel_weight') def construct(self, input_, label): input_n = self.normalize(input_) w = self.normalize2(self.weight) fc_o = self.fc(input_n, w) fc_o_shape = F.shape(fc_o) one_hot_float = self.onehot(label, fc_o_shape[1], self.on_value, self.off_value) local_label = self.cast(one_hot_float, mstype.int32) exp_o = self.exp(fc_o) mul_const_o = self.mul_const(self.a_const, exp_o) mul_const2_o = self.mul_const2(self.b_const, mul_const_o) exp2_o = self.exp2(mul_const2_o) mul_const3_o = self.mul_const3(exp2_o, self.c_const) mul_const4_o = self.mul_const4(F.scalar_to_array(1), local_label) mul6_o = self.mul6(self.mul(mul_const3_o, one_hot_float), self.mul2(fc_o, self.cast2(mul_const4_o, mstype.float32))) mul_const5_o = self.mul_const5(mul6_o, self.d_const) max_o = self.reduce_max(mul_const5_o, -1) mul4_o = self.mul4(mul_const5_o, max_o) exp3_o = self.exp3(mul4_o) sum_o = self.reduce_sum(exp3_o, -1) reshape_o = self.reshape(sum_o, (F.shape(sum_o)[0], 1)) mul5_o = self.mul5(exp3_o, reshape_o) log_o = self.log(self.mul9(mul5_o, self.e_const)) mul3_o = self.mul3(log_o, one_hot_float) mul7_o = self.mul7(mul3_o, self.cast3(F.scalar_to_array(-1), mstype.float32)) sum2_o = self.reduce_sum_2(mul7_o, -1) loss = self.mul8(self.reduce_sum_3(sum2_o, -1), self.cast4(F.scalar_to_array(F.shape(mul_const5_o)[0]), mstype.float32)) return loss class Dataset(MindData): def __init__(self, predict, label, length=3, input_num=2): super(Dataset, self).__init__(size=length) self.predict = predict self.label = label self.index = 0 self.length = length self.input_num = input_num def __iter__(self): return self def __next__(self): if self.index >= self.length: raise StopIteration self.index += 1 if self.input_num == 2: return (self.predict, self.label) return (self.predict,) def reset(self): self.index = 0 class NetWithLoss(nn.Cell): def __init__(self, network): super(NetWithLoss, self).__init__() self.loss = VirtualLoss() self.network = network def construct(self, x, b): predict = self.network(x, b) return self.loss(predict) class GradWrap(nn.Cell): def __init__(self, network): super(GradWrap, self).__init__() self.network = network def construct(self, x, b): return C.grad_all(self.network)(x, b) def bn_with_initialize(out_channels): bn = nn.BatchNorm2d(out_channels, momentum=0.3, eps=1e-5).add_flags_recursive(fp32=True) return bn def fc_with_initialize(input_channels, out_channels): return nn.Dense(input_channels, out_channels) class BNReshapeDenseBNNet(nn.Cell): def __init__(self): super(BNReshapeDenseBNNet, self).__init__() self.batch_norm = bn_with_initialize(2) self.reshape = P.Reshape() self.batch_norm2 = nn.BatchNorm1d(512, affine=False) self.fc = fc_with_initialize(2 * 32 * 32, 512) self.loss = SemiAutoOneHotNet(args=Args(), strategy=StrategyBatch()) def construct(self, x, label): x = self.batch_norm(x) x = self.reshape(x, (16, 2 * 32 * 32)) x = self.fc(x) x = self.batch_norm2(x) loss = self.loss(x, label) return loss def test_bn_reshape_dense_bn_train_loss(): batch_size = 16 context.set_auto_parallel_context(device_num=device_num, global_rank=0) input_ = Tensor(np.ones([batch_size, 2, 32, 32]).astype(np.float32) * 0.01) label = Tensor(np.ones([batch_size]), dtype=ms.int32) net = GradWrap(NetWithLoss(BNReshapeDenseBNNet())) context.set_auto_parallel_context(parallel_mode="semi_auto_parallel") net.set_auto_parallel() _executor.compile(net, input_, label) def test_semi_one_hot_net_batch(): batch_size = 16 context.set_auto_parallel_context(device_num=device_num, global_rank=0) input_ = Tensor(np.ones([batch_size * 1, 512]).astype(np.float32) * 0.01) label = Tensor(np.ones([batch_size]), dtype=ms.int32) net = SemiAutoOneHotNet(args=Args(), strategy=StrategyBatch()) net = GradWrap(NetWithLoss(net)) context.set_auto_parallel_context(parallel_mode="semi_auto_parallel") net.set_auto_parallel() _executor.compile(net, input_, label) def test_semi_one_hot_net_model(): batch_size = 16 learning_rate = 0.1 momentum = 0.9 epoch_size = 2 predict = Tensor(np.ones([batch_size, 512]), dtype=ms.float32) label = Tensor(np.ones([batch_size]), dtype=ms.int32) dataset = Dataset(predict, label, 2, input_num=2) net = SemiAutoOneHotNet(args=Args(), strategy=StrategyModel()) opt = Momentum(net.trainable_params(), learning_rate, momentum) context.reset_auto_parallel_context() context.set_auto_parallel_context(parallel_mode=ParallelMode.SEMI_AUTO_PARALLEL, device_num=16) context.set_context(mode=context.GRAPH_MODE) model = Model(net, optimizer=opt) model.train(epoch_size, dataset, dataset_sink_mode=False)