Merge pull request #480 from sony/feature/20190701-model-auto-forward…

…-support

[Models] auto-forward support and modified parameter loading method.

python/src/nnabla/utils/nnp_graph.py

@@ -274,20 +274,34 @@ def _get_variable_or_create(self, v, callback, current_scope):
             var.name = name
             return var
 
-        # Trying to load the parameter from .nnp file.
-        callback.verbose(
-            'Loading parameter `{}` from .nnp.'.format(name))
+        # Trying to load the parameter from the global scope.
         try:
+            with nn.parameter_scope('', current_scope):
+                param = get_parameter(name)
+
+            if param is not None:
+                assert shape == param.shape
+                param = param.get_unlinked_variable(need_grad=v.need_grad)
+                v.variable = param
+                param.name = name
+                return param
+
+            # Parameter does not exist in the global scope.
+            # Then try to load the parameter from .nnp file.
+            callback.verbose(
+                'Loading parameter `{}` from .nnp.'.format(name))
             param = get_parameter(name)
+
             if param is None:
                 logger.info(
                     'Parameter `{}` is not found. Initializing.'.format(name))
                 tmp = _create_variable(pvar, name, shape, self.rng)
                 param = tmp.variable_instance
-                set_parameter(name, param)
-            # Always copy param to current scope even if it already exists.
+
+            # Register the parameter to the current (global) scope.
             with nn.parameter_scope('', current_scope):
                 set_parameter(name, param)
+
         except:
             import sys
             import traceback
@@ -298,6 +312,7 @@ def _get_variable_or_create(self, v, callback, current_scope):
                     name, traceback.format_exc(),
                     '\n'.join(
                         list(nn.get_parameters(grad_only=False).keys()))))
+
         assert shape == param.shape
         param = param.get_unlinked_variable(need_grad=v.need_grad)
         v.variable = param
@@ -321,7 +336,7 @@ def _create_function(self, f, callback, current_scope):
         function_instance = _create_function(inputs, f.proto, self.batch_size)
 
         outputs = function_instance(
-            *inputs, n_outputs=len(f.outputs), auto_forward=False)
+            *inputs, n_outputs=len(f.outputs), auto_forward=nn.get_auto_forward())
         if not isinstance(outputs, tuple):
             outputs = (outputs,)