!451 recitify resnet50 performance

Merge pull request !451 from gengdongjie/r0.5

tutorials/tutorial_code/sample_for_cloud/resnet50_train.py

@@ -27,6 +27,7 @@ from mindspore.nn.loss import SoftmaxCrossEntropyWithLogits
 from mindspore.train.model import Model, ParallelMode
 from mindspore.train.callback import Callback, LossMonitor
 from mindspore.train.loss_scale_manager import FixedLossScaleManager
+from mindspore.communication.management import init
 import mindspore.dataset.engine as de
 
 from dataset import create_dataset, device_id, device_num
@@ -121,6 +122,7 @@ def resnet50_train(args_opt):
         context.set_auto_parallel_context(device_num=device_num,
                                           parallel_mode=ParallelMode.DATA_PARALLEL,
                                           mirror_mean=True)
+        init()
         local_data_path = os.path.join(local_data_path, str(device_id))
 
     # data download
@@ -138,12 +140,12 @@ def resnet50_train(args_opt):
 
     # create model
     net = resnet50(class_num = class_num)
-    loss = SoftmaxCrossEntropyWithLogits(sparse=True)
+    loss = SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
     lr = Tensor(get_lr(global_step=0, total_epochs=epoch_size, steps_per_epoch=train_step_size))
     opt = Momentum(net.trainable_params(), lr, momentum=0.9, weight_decay=1e-4, loss_scale=loss_scale_num)
     loss_scale = FixedLossScaleManager(loss_scale_num, False)
 
-    model = Model(net, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale, metrics={'acc'})
+    model = Model(net, amp_level="O2", keep_batchnorm_fp32=False, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale, metrics={'acc'})
 
     # define performance callback to show ips and loss callback to show loss for every epoch
     performance_cb = PerformanceCallback(batch_size)