!1510 seperate auto_parallel and stand_alone when init initializer data

Merge pull request !1510 from yihuaijie/dev

mindspore/common/api.py

@@ -327,16 +327,19 @@ class _Executor:
             raise TypeError('Parameters need OrderedDict type, but got {}'.
                             format(type(params)))
 
-    def _params_init_data(self, obj, params):
+    def _params_init_data(self, obj, params, auto_parallel_mode=False):
+        """Init parameters' data."""
         if params is not None:
             for key, param in params.items():
-                if key not in obj.parameter_layout_dict:
-                    logger.info("Layout dict does not contain the key %s.", key)
+                if not auto_parallel_mode:
                     param.init_data()
+                elif key not in obj.parameter_layout_dict:
+                    logger.info("Layout dict does not contain the key %s.", key)
+                    param.init_data(set_sliced=True)
                 else:
                     layout = obj.parameter_layout_dict[key]
-                    param.init_data(layout)
-        obj.init_parameters_data()
+                    param.init_data(layout, set_sliced=True)
+        obj.init_parameters_data(auto_parallel_mode=auto_parallel_mode)
 
     def compile(self, obj, *args, phase='predict', params=None, do_convert=True, auto_parallel_mode=False):
         """
@@ -383,11 +386,11 @@ class _Executor:
         if not do_convert:
             return phase, True
 
-        if auto_parallel_mode and "train" in phase:
+        if auto_parallel_mode:
             obj.parameter_layout_dict = self._executor.get_parameter_layout(phase)
-        self._params_init_data(obj, params)
+        self._params_init_data(obj, params, auto_parallel_mode)
         if not enable_debug_runtime or enable_ge:
-            if auto_parallel_mode and "train" in phase:
+            if auto_parallel_mode:
                 obj.load_parameter_slice(params)
 
         # set parallel inputs in sink mode

mindspore/common/parameter.py

@@ -99,6 +99,10 @@ class Parameter:
         """Get slice status of the parameter."""
         return self._sliced
 
+    @sliced.setter
+    def sliced(self, sliced_):
+        self._sliced = sliced_
+
     @property
     def is_init(self):
         """Get init status of the parameter."""
@@ -211,15 +215,18 @@ class Parameter:
         self.default_input = data
 
 
-    def init_data(self, layout=None):
+    def init_data(self, layout=None, set_sliced=False):
         """
         Init data of the parameter.
 
         Args:
-            layout (list[list[int]]): parameter slice layout [dev_mat, tensor_map, slice_shape].
-                dev_mat (list[int]): device matrix.
-                tensor_map (list[int]): tensor map.
-                slice_shape (list[int]): shape of slice.
+            layout (list[list[int]]): Parameter slice layout [dev_mat, tensor_map, slice_shape].
+
+                - dev_mat (list[int]): Device matrix.
+                - tensor_map (list[int]): Tensor map.
+                - slice_shape (list[int]): Shape of slice.
+            set_sliced (bool): True if should set parameter sliced after init the data of initializer.
+                Default: False.
         """
         if not isinstance(self.default_input, MetaTensor):
             return
@@ -235,7 +242,8 @@ class Parameter:
 
         self.default_input = self.init_mode.to_tensor()
         self.init_mode = None
-        self._sliced = True
+        if set_sliced:
+            self.sliced = True
 
 
 class ParameterTuple(tuple):

mindspore/nn/cell.py

@@ -264,11 +264,12 @@ class Cell:
                     logger.info("layout dict does not contain the key %s", key)
                     continue
                 if self.parameters_dict()[key].sliced:
-                    logger.info("Param %s is from initializer, already sliced.", key)
+                    logger.info("Param %s is already sliced.", key)
                     continue
                 layout = self.parameter_layout_dict[key]
                 new_tensor = _load_tensor_by_layout(tensor, layout)
                 self.parameters_dict()[key].set_parameter_data(new_tensor)
+                self.parameters_dict()[key].sliced = True
         elif isinstance(params, OrderedDict):
             for key in params:
                 tensor = params[key].data
@@ -276,11 +277,12 @@ class Cell:
                     logger.info("layout dict does not contain the key %s", key)
                     continue
                 if params[key].sliced:
-                    logger.info("Param %s is from initializer, already sliced.", key)
+                    logger.info("Param %s is already sliced.", key)
                     continue
                 layout = self.parameter_layout_dict[key]
                 new_tensor = _load_tensor_by_layout(tensor, layout)
                 params[key].set_parameter_data(new_tensor)
+                params[key].sliced = True
         else:
             raise TypeError('Parameters need OrderedDict type, but got {}'.
                             format(type(params)))
@@ -435,14 +437,17 @@ class Cell:
         """
         raise NotImplementedError
 
-    def init_parameters_data(self, recurse=True):
+    def init_parameters_data(self, recurse=True, auto_parallel_mode=False):
+        """Init parameters' data."""
         for param in self.get_parameters(expand=recurse):
-            if param.name not in self.parameter_layout_dict:
-                logger.info("Layout dict does not contain the key %s.", param.name)
+            if not auto_parallel_mode:
                 param.init_data()
+            elif param.name not in self.parameter_layout_dict:
+                logger.info("Layout dict does not contain the key %s.", param.name)
+                param.init_data(set_sliced=True)
             else:
                 layout = self.parameter_layout_dict[param.name]
-                param.init_data(layout)
+                param.init_data(layout, set_sliced=True)
 
     def parameters_dict(self, recurse=True):
         """