rename mirror_mean to gradients_mean

mindspore/ccsrc/frontend/parallel/context.cc

@@ -45,7 +45,7 @@ std::shared_ptr<ParallelContext> ParallelContext::GetInstance() {
 ParallelContext::ParallelContext() { Reset(); }
 
 void ParallelContext::Reset() {
-  mirror_mean_ = false;
+  gradients_mean_ = false;
   full_batch_ = false;
   gradient_fp32_sync_ = true;
   loss_repeated_mean_ = true;
@@ -74,7 +74,7 @@ void ParallelContext::set_global_rank(int32_t global_rank) {
   global_rank_is_set_ = true;
 }
 
-void ParallelContext::set_mirror_mean(bool mirror_mean) { mirror_mean_ = mirror_mean; }
+void ParallelContext::set_gradients_mean(bool gradients_mean) { gradients_mean_ = gradients_mean; }
 
 void ParallelContext::set_full_batch(bool full_batch) { full_batch_ = full_batch; }
 

mindspore/ccsrc/frontend/parallel/context.h

@@ -52,8 +52,8 @@ class ParallelContext {
 
   static std::shared_ptr<ParallelContext> GetInstance();
 
-  void set_mirror_mean(bool mirror_mean);
-  bool mirror_mean() const { return mirror_mean_; }
+  void set_gradients_mean(bool gradients_mean);
+  bool gradients_mean() const { return gradients_mean_; }
 
   void set_full_batch(bool full_batch);
   bool full_batch() const { return full_batch_; }
@@ -107,7 +107,7 @@ class ParallelContext {
  private:
   ParallelContext();
   static std::shared_ptr<ParallelContext> inst_context_;
-  bool mirror_mean_;
+  bool gradients_mean_;
   bool full_batch_;
   bool gradient_fp32_sync_;
   bool loss_repeated_mean_;

mindspore/ccsrc/frontend/parallel/ops_info/operator_info.cc

@@ -251,7 +251,7 @@ OperatorVector CreateMirrorOps(const std::string &group_name, size_t dev_num) {
     MS_LOG(EXCEPTION) << "Invalid dev num: " << dev_num;
   }
   OperatorVector op_for_weight;
-  bool mean_flag = ParallelContext::GetInstance()->mirror_mean();
+  bool mean_flag = ParallelContext::GetInstance()->gradients_mean();
 
   OperatorName operator_name = MIRROR_OPERATOR;
   ValuePtr attr0_value = MakeValue(group_name);

mindspore/ccsrc/frontend/parallel/step_parallel.cc

@@ -2488,7 +2488,7 @@ Status ParallelInit() {
   }
 
   MS_LOG(INFO) << "The parallel context: dev num: " << device_num << ", global rank: " << global_rank
-               << ", backend: " << backend << ", mirror_mean: " << ParallelContext::GetInstance()->mirror_mean()
+               << ", backend: " << backend << ", gradients_mean: " << ParallelContext::GetInstance()->gradients_mean()
                << ", gradient_fp32_sync: " << ParallelContext::GetInstance()->gradient_fp32_sync();
   return SUCCESS;
 }

mindspore/ccsrc/pipeline/jit/init.cc

@@ -113,8 +113,8 @@ PYBIND11_MODULE(_c_expression, m) {
     .def("get_global_rank", &ParallelContext::global_rank, "Get global rank.")
     .def("set_global_rank", &ParallelContext::set_global_rank, "Set global rank.")
     .def("get_global_rank_is_set", &ParallelContext::global_rank_is_set, "Get global rank is set.")
-    .def("get_mirror_mean", &ParallelContext::mirror_mean, "Get mirror mean.")
-    .def("set_mirror_mean", &ParallelContext::set_mirror_mean, "Set mirror mean.")
+    .def("get_gradients_mean", &ParallelContext::gradients_mean, "Get mirror mean.")
+    .def("set_gradients_mean", &ParallelContext::set_gradients_mean, "Set mirror mean.")
     .def("get_gradient_fp32_sync", &ParallelContext::gradient_fp32_sync, "Get cast before mirror.")
     .def("set_gradient_fp32_sync", &ParallelContext::set_gradient_fp32_sync, "Set cast before mirror.")
     .def("get_loss_repeated_mean", &ParallelContext::loss_repeated_mean, "Get loss repeated mean.")

mindspore/context.py

@@ -323,7 +323,7 @@ def _context():
     return _k_context
 
 
-@args_type_check(device_num=int, global_rank=int, mirror_mean=bool, gradient_fp32_sync=bool, parallel_mode=str,
+@args_type_check(device_num=int, global_rank=int, gradients_mean=bool, gradient_fp32_sync=bool, parallel_mode=str,
                  auto_parallel_search_mode=str, parameter_broadcast=bool, strategy_ckpt_load_file=str,
                  strategy_ckpt_save_file=str, full_batch=bool, enable_parallel_optimizer=bool)
 def set_auto_parallel_context(**kwargs):
@@ -341,8 +341,8 @@ def set_auto_parallel_context(**kwargs):
     Args:
         device_num (int): Available device number, the value must be in [1, 4096]. Default: 1.
         global_rank (int): Global rank id, the value must be in [0, 4095]. Default: 0.
-        mirror_mean (bool): Whether to perform mean operator after all-reduce of mirror.
-                     "stand_alone" do not support mirror_mean. Default: False.
+        gradients_mean (bool): Whether to perform mean operator after all-reduce of mirror.
+                     "stand_alone" do not support gradients_mean. Default: False.
         gradient_fp32_sync (bool): Gradients allreduce by fp32 even though gradients is fp16 if this flag is True..
                      "stand_alone", "data_parallel" and "hybrid_parallel" do not support
                      gradient_fp32_sync. Default: True.
@@ -380,7 +380,7 @@ def set_auto_parallel_context(**kwargs):
     Examples:
         >>> context.set_auto_parallel_context(device_num=8)
         >>> context.set_auto_parallel_context(global_rank=0)
-        >>> context.set_auto_parallel_context(mirror_mean=True)
+        >>> context.set_auto_parallel_context(gradients_mean=True)
         >>> context.set_auto_parallel_context(gradient_fp32_sync=False)
         >>> context.set_auto_parallel_context(parallel_mode="auto_parallel")
         >>> context.set_auto_parallel_context(parameter_broadcast=False)
@@ -412,7 +412,7 @@ def reset_auto_parallel_context():
 
     - device_num: 1.
     - global_rank: 0.
-    - mirror_mean: False.
+    - gradients_mean: False.
     - gradient_fp32_sync: True.
     - parallel_mode: "stand_alone".
     - parameter_broadcast: False.

mindspore/nn/wrap/cell_wrapper.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 # ============================================================================
 """Cell_wrapper."""
-from mindspore.parallel._utils import (_get_device_num, _get_mirror_mean,
+from mindspore.parallel._utils import (_get_device_num, _get_gradients_mean,
                                        _get_parallel_mode)
 from mindspore.context import ParallelMode
 from ...common import dtype as mstype
@@ -190,7 +190,7 @@ class TrainOneStepCell(Cell):
         if parallel_mode in (ParallelMode.DATA_PARALLEL, ParallelMode.HYBRID_PARALLEL):
             self.reducer_flag = True
         if self.reducer_flag:
-            mean = _get_mirror_mean()
+            mean = _get_gradients_mean()
             degree = _get_device_num()
             self.grad_reducer = DistributedGradReducer(optimizer.parameters, mean, degree)
 

mindspore/nn/wrap/grad_reducer.py

@@ -279,7 +279,7 @@ class DistributedGradReducer(Cell):
         >>>                                            ParallelMode.HYBRID_PARALLEL]:
         >>>             self.reducer_flag = True
         >>>         if self.reducer_flag:
-        >>>             mean = context.get_auto_parallel_context("mirror_mean")
+        >>>             mean = context.get_auto_parallel_context("gradients_mean")
         >>>             if mean.get_device_num_is_set():
         >>>                 degree = context.get_auto_parallel_context("device_num")
         >>>             else:

mindspore/nn/wrap/loss_scale.py

@@ -16,7 +16,7 @@
 import mindspore.context as context
 from mindspore.nn.wrap.grad_reducer import DistributedGradReducer
 from mindspore.context import ParallelMode
-from mindspore.parallel._utils import _get_device_num, _get_parallel_mode, _get_mirror_mean
+from mindspore.parallel._utils import _get_device_num, _get_parallel_mode, _get_gradients_mean
 from ..cell import Cell
 from ...common import Tensor, RowTensor
 from ...common.parameter import Parameter
@@ -231,7 +231,7 @@ class TrainOneStepWithLossScaleCell(Cell):
         self.grad_reducer = F.identity
         self.reducer_flag = self.parallel_mode in [ParallelMode.DATA_PARALLEL, ParallelMode.HYBRID_PARALLEL]
         if self.reducer_flag:
-            mean = _get_mirror_mean()
+            mean = _get_gradients_mean()
             degree = _get_device_num()
             self.grad_reducer = DistributedGradReducer(optimizer.parameters, mean, degree)
         self.is_distributed = self.parallel_mode != ParallelMode.STAND_ALONE

mindspore/parallel/_auto_parallel_context.py

@@ -95,23 +95,23 @@ class _AutoParallelContext:
         self.check_context_handle()
         return self._context_handle.get_global_rank()
 
-    def set_mirror_mean(self, mirror_mean):
+    def set_gradients_mean(self, gradients_mean):
         """
-        Set mirror_mean flag.
+        Set gradients_mean flag.
 
         Note:
-            If mirror_mean is true, it will insert a div operator after parameter gradients allreduce.
+            If gradients_mean is true, it will insert a div operator after parameter gradients allreduce.
 
         Args:
-            mirror_mean (bool): The mirror_mean flag.
+            gradients_mean (bool): The gradients_mean flag.
         """
         self.check_context_handle()
-        self._context_handle.set_mirror_mean(mirror_mean)
+        self._context_handle.set_gradients_mean(gradients_mean)
 
-    def get_mirror_mean(self):
-        """Get mirror_mean flag."""
+    def get_gradients_mean(self):
+        """Get gradients_mean flag."""
         self.check_context_handle()
-        return self._context_handle.get_mirror_mean()
+        return self._context_handle.get_gradients_mean()
 
     def set_gradient_fp32_sync(self, gradient_fp32_sync):
         """
@@ -453,7 +453,7 @@ def auto_parallel_context():
 _set_auto_parallel_context_func_map = {
     "device_num": auto_parallel_context().set_device_num,
     "global_rank": auto_parallel_context().set_global_rank,
-    "mirror_mean": auto_parallel_context().set_mirror_mean,
+    "gradients_mean": auto_parallel_context().set_gradients_mean,
     "gradient_fp32_sync": auto_parallel_context().set_gradient_fp32_sync,
     "loss_repeated_mean": auto_parallel_context().set_loss_repeated_mean,
     "parallel_mode": auto_parallel_context().set_parallel_mode,
@@ -468,7 +468,7 @@ _set_auto_parallel_context_func_map = {
 _get_auto_parallel_context_func_map = {
     "device_num": auto_parallel_context().get_device_num,
     "global_rank": auto_parallel_context().get_global_rank,
-    "mirror_mean": auto_parallel_context().get_mirror_mean,
+    "gradients_mean": auto_parallel_context().get_gradients_mean,
     "gradient_fp32_sync": auto_parallel_context().get_gradient_fp32_sync,
     "loss_repeated_mean": auto_parallel_context().get_loss_repeated_mean,
     "parallel_mode": auto_parallel_context().get_parallel_mode,
@@ -480,7 +480,7 @@ _get_auto_parallel_context_func_map = {
     "enable_parallel_optimizer": auto_parallel_context().get_enable_parallel_optimizer}
 
 
-@args_type_check(device_num=int, global_rank=int, mirror_mean=bool, gradient_fp32_sync=bool,
+@args_type_check(device_num=int, global_rank=int, gradients_mean=bool, gradient_fp32_sync=bool,
                  loss_repeated_mean=bool, parallel_mode=str, auto_parallel_search_mode=str,
                  parameter_broadcast=bool, strategy_ckpt_load_file=str,
                  strategy_ckpt_save_file=str, full_batch=bool, enable_parallel_optimizer=bool)
@@ -495,7 +495,7 @@ def _set_auto_parallel_context(**kwargs):
     Args:
         device_num (int): Available device number, the value must be in [1, 4096]. Default: 1.
         global_rank (int): Global rank id, the value must be in [0, 4095]. Default: 0.
-        mirror_mean (bool): Whether to perform mean operator after all-reduce of mirror. Default: False.
+        gradients_mean (bool): Whether to perform mean operator after all-reduce of mirror. Default: False.
         loss_repeated_mean (bool): Whether to perform mean operator in backward in the case of repeated
                         calculations. Default: True.
         gradient_fp32_sync (bool): Gradients allreduce by fp32 even though gradients is fp16 if this flag is True.
@@ -562,7 +562,7 @@ def _reset_auto_parallel_context():
 
     - device_num: 1.
     - global_rank: 0.
-    - mirror_mean: False.
+    - gradients_mean: False.
     - gradient_fp32_sync: True.
     - parallel_mode: "stand_alone".
     - parameter_broadcast: False.

mindspore/parallel/_utils.py

@@ -88,9 +88,9 @@ def _to_full_tensor(elem, device_num, global_rank, scaling_sens=None):
         lst.append(Tensor(scaling_sens, mstype.float32))
     return tuple(lst)
 
-def _get_mirror_mean():
-    """Get if using mirror_mean."""
-    return auto_parallel_context().get_mirror_mean()
+def _get_gradients_mean():
+    """Get if using gradients_mean."""
+    return auto_parallel_context().get_gradients_mean()
 
 
 def _get_device_num():

model_zoo/official/cv/deeplabv3/train.py

@@ -66,7 +66,7 @@ def model_fine_tune(flags, train_net, fix_weight_layer):
             para.requires_grad = False
 if __name__ == "__main__":
     if args_opt.distribute == "true":
-        context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL, mirror_mean=True)
+        context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL, gradients_mean=True)
         init()
     args_opt.base_size = config.crop_size
     args_opt.crop_size = config.crop_size

model_zoo/official/cv/faster_rcnn/train.py

@@ -54,7 +54,7 @@ if __name__ == '__main__':
         rank = args_opt.rank_id
         device_num = args_opt.device_num
         context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
-                                          mirror_mean=True, parameter_broadcast=True)
+                                          gradients_mean=True, parameter_broadcast=True)
         init()
     else:
         rank = 0

model_zoo/official/cv/googlenet/train.py

@@ -78,7 +78,7 @@ if __name__ == '__main__':
         if device_num > 1:
             context.reset_auto_parallel_context()
             context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
-                                              mirror_mean=True)
+                                              gradients_mean=True)
             init()
     elif device_target == "GPU":
         init()
@@ -86,7 +86,7 @@ if __name__ == '__main__':
         if device_num > 1:
             context.reset_auto_parallel_context()
             context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
-                                              mirror_mean=True)
+                                              gradients_mean=True)
     else:
         raise ValueError("Unsupported platform.")
 

model_zoo/official/cv/inceptionv3/train.py

@@ -58,7 +58,7 @@ if __name__ == '__main__':
         cfg.group_size = get_group_size()
         parallel_mode = ParallelMode.DATA_PARALLEL
         context.set_auto_parallel_context(parallel_mode=parallel_mode, device_num=cfg.group_size,
-                                          parameter_broadcast=True, mirror_mean=True)
+                                          parameter_broadcast=True, gradients_mean=True)
     else:
         cfg.rank = 0
         cfg.group_size = 1

model_zoo/official/cv/maskrcnn/train.py

@@ -58,7 +58,7 @@ if __name__ == '__main__':
         rank = args_opt.rank_id
         device_num = args_opt.device_num
         context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
-                                          mirror_mean=True, parameter_broadcast=True)
+                                          gradients_mean=True, parameter_broadcast=True)
         init()
     else:
         rank = 0

model_zoo/official/cv/mobilenetv2/src/utils.py

@@ -39,7 +39,7 @@ def context_device_init(config):
         init("nccl")
         context.set_auto_parallel_context(device_num=get_group_size(),
                                           parallel_mode=ParallelMode.DATA_PARALLEL,
-                                          mirror_mean=True)
+                                          gradients_mean=True)
 
     elif config.platform == "Ascend":
         context.set_context(mode=context.GRAPH_MODE, device_target=config.platform, device_id=config.device_id,
@@ -47,7 +47,7 @@ def context_device_init(config):
         if config.run_distribute:
             context.set_auto_parallel_context(device_num=config.rank_size,
                                               parallel_mode=ParallelMode.DATA_PARALLEL,
-                                              parameter_broadcast=True, mirror_mean=True)
+                                              parameter_broadcast=True, gradients_mean=True)
             auto_parallel_context().set_all_reduce_fusion_split_indices([140])
             init()
     else:

model_zoo/official/cv/mobilenetv2_quant/train.py

@@ -57,7 +57,7 @@ elif args_opt.device_target == "GPU":
     init()
     context.set_auto_parallel_context(device_num=get_group_size(),
                                       parallel_mode=ParallelMode.DATA_PARALLEL,
-                                      mirror_mean=True)
+                                      gradients_mean=True)
     context.set_context(mode=context.GRAPH_MODE,
                         device_target="GPU",
                         save_graphs=False)
@@ -77,7 +77,7 @@ def train_on_ascend():
         context.set_auto_parallel_context(device_num=rank_size,
                                           parallel_mode=ParallelMode.DATA_PARALLEL,
                                           parameter_broadcast=True,
-                                          mirror_mean=True)
+                                          gradients_mean=True)
         init()
 
     # define network

model_zoo/official/cv/mobilenetv3/train.py

@@ -55,7 +55,7 @@ if args_opt.device_target == "GPU":
     init()
     context.set_auto_parallel_context(device_num=get_group_size(),
                                       parallel_mode=ParallelMode.DATA_PARALLEL,
-                                      mirror_mean=True)
+                                      gradients_mean=True)
 else:
     raise ValueError("Unsupported device_target.")
 

model_zoo/official/cv/nasnet/src/nasnet_a_mobile.py

@@ -24,7 +24,7 @@ import mindspore.ops.composite as C
 import mindspore.common.dtype as mstype
 from mindspore.nn.wrap.grad_reducer import DistributedGradReducer
 from mindspore.train.parallel_utils import ParallelMode
-from mindspore.parallel._utils import _get_device_num, _get_parallel_mode, _get_mirror_mean
+from mindspore.parallel._utils import _get_device_num, _get_parallel_mode, _get_gradients_mean
 
 
 GRADIENT_CLIP_TYPE = 1
@@ -921,7 +921,7 @@ class NASNetAMobileTrainOneStepWithClipGradient(nn.Cell):
         if parallel_mode in (ParallelMode.DATA_PARALLEL, ParallelMode.HYBRID_PARALLEL):
             self.reducer_flag = True
         if self.reducer_flag:
-            mean = _get_mirror_mean()
+            mean = _get_gradients_mean()
             degree = _get_device_num()
             self.grad_reducer = DistributedGradReducer(optimizer.parameters, mean, degree)
 

model_zoo/official/cv/nasnet/train.py

@@ -58,7 +58,7 @@ if __name__ == '__main__':
         cfg.group_size = get_group_size()
         parallel_mode = ParallelMode.DATA_PARALLEL
         context.set_auto_parallel_context(parallel_mode=parallel_mode, device_num=cfg.group_size,
-                                          parameter_broadcast=True, mirror_mean=True)
+                                          parameter_broadcast=True, gradients_mean=True)
     else:
         cfg.rank = 0
         cfg.group_size = 1

model_zoo/official/cv/resnet/train.py

@@ -76,7 +76,7 @@ if __name__ == '__main__':
             device_id = int(os.getenv('DEVICE_ID'))
             context.set_context(device_id=device_id, enable_auto_mixed_precision=True)
             context.set_auto_parallel_context(device_num=args_opt.device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
-                                              mirror_mean=True)
+                                              gradients_mean=True)
             if args_opt.net == "resnet50" or args_opt.net == "se-resnet50":
                 auto_parallel_context().set_all_reduce_fusion_split_indices([85, 160])
             else:
@@ -86,7 +86,7 @@ if __name__ == '__main__':
         else:
             init()
             context.set_auto_parallel_context(device_num=get_group_size(), parallel_mode=ParallelMode.DATA_PARALLEL,
-                                              mirror_mean=True)
+                                              gradients_mean=True)
             if args_opt.net == "resnet50":
                 auto_parallel_context().set_all_reduce_fusion_split_indices([85, 160])
             ckpt_save_dir = config.save_checkpoint_path + "ckpt_" + str(get_rank()) + "/"

model_zoo/official/cv/resnet50_quant/train.py

@@ -76,11 +76,11 @@ if __name__ == '__main__':
         context.set_auto_parallel_context(device_num=rank_size,
                                           parallel_mode=ParallelMode.DATA_PARALLEL,
                                           parameter_broadcast=True,
-                                          mirror_mean=True)
+                                          gradients_mean=True)
         init()
         context.set_auto_parallel_context(device_num=args_opt.device_num,
                                           parallel_mode=ParallelMode.DATA_PARALLEL,
-                                          mirror_mean=True)
+                                          gradients_mean=True)
         auto_parallel_context().set_all_reduce_fusion_split_indices([107, 160])
 
     # define network

model_zoo/official/cv/resnet_thor/src/grad_reducer_thor.py

@@ -129,7 +129,7 @@ class DistributedGradReducerThor(Cell):
         >>>                                            ParallelMode.HYBRID_PARALLEL]:
         >>>             self.reducer_flag = True
         >>>         if self.reducer_flag:
-        >>>             mean = context.get_auto_parallel_context("mirror_mean")
+        >>>             mean = context.get_auto_parallel_context("gradients_mean")
         >>>             if mean.get_device_num_is_set():
         >>>                 degree = context.get_auto_parallel_context("device_num")
         >>>             else:

model_zoo/official/cv/resnet_thor/src/thor.py

@@ -22,7 +22,7 @@ import mindspore.common.dtype as mstype
 from mindspore._checkparam import check_bool
 from mindspore._checkparam import Validator as validator
 from mindspore.nn.optim.optimizer import Optimizer
-from mindspore.parallel._utils import _get_device_num, _get_mirror_mean
+from mindspore.parallel._utils import _get_device_num, _get_gradients_mean
 from src.grad_reducer_thor import DistributedGradReducerThor
 
 _momentum_opt = C.MultitypeFuncGraph("momentum_opt")
@@ -85,7 +85,7 @@ class THOR_GPU(Optimizer):
         self.assign = P.Assign()
         self.mul = P.Mul()
 
-        mean = _get_mirror_mean()
+        mean = _get_gradients_mean()
         degree = _get_device_num()
         self.grad_reducer_thorA = DistributedGradReducerThor(self.parameters, 0, mean, degree)
         self.grad_reducer_thorG = DistributedGradReducerThor(self.parameters, 0, mean, degree)
@@ -191,7 +191,7 @@ class THOR(Optimizer):
                             1.0 / 196, 1.0 / 196, 1.0 / 196,
                             1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49,
                             1.0]
-        mean = _get_mirror_mean()
+        mean = _get_gradients_mean()
         degree = _get_device_num()
         self.grad_reducer_Amax = DistributedGradReducerThor(self.parameters, 2, mean, degree)
         self.grad_reducer_Gmax = DistributedGradReducerThor(self.parameters, 5, mean, degree)