Yj/test branch

python/paddle/common_ops_import.py

@@ -24,7 +24,6 @@
     OpProtoHolder,
     Variable,
     _dygraph_tracer,
-    _in_legacy_dygraph,
     _non_static_mode,
     _varbase_creator,
     convert_np_dtype_to_dtype_,

python/paddle/fluid/dygraph/tracer.py

@@ -306,7 +306,7 @@ def trace_op(
         stop_gradient=False,
         inplace_map=None,
     ):
-        if not framework._in_legacy_dygraph():
+        if framework.in_dygraph_mode():
             # inputs : {"sum": [tensor], ...}
             # outputs : {"sum": [tensor], ...}
             if type in name_mapping.keys():

python/paddle/fluid/framework.py

@@ -237,9 +237,6 @@ def in_dygraph_mode():
     return (_dygraph_tracer_ is not None) and _in_eager_mode_
 
 
-def _in_legacy_dygraph():
-    return (not _in_eager_mode_) and (_dygraph_tracer_ is not None)
-
 
 def _non_static_mode():
     return _dygraph_tracer_ is not None
@@ -255,7 +252,8 @@ def _test_eager_guard(place=None):
     try:
         yield
     finally:
-        pass
+        if not already_fallback:
+            _enable_legacy_dygraph()
 
 
 global_ipu_index = -1
@@ -1334,8 +1332,6 @@ def __instancecheck__(cls, instance):
         if in_dygraph_mode():
             return issubclass(t, core.eager.Tensor)
         else:
-            if _in_legacy_dygraph():
-                return issubclass(t, core.VarBase)
             return issubclass(t, Variable)
 
 
@@ -1346,8 +1342,6 @@ def __instancecheck__(cls, instance):
         if in_dygraph_mode():
             return issubclass(t, EagerParamBase)
         else:
-            if _in_legacy_dygraph():
-                return issubclass(t, ParamBase)
             return issubclass(t, Parameter)
 
 
@@ -3893,19 +3887,6 @@ def _rename_var(self, name, new_name):
                     error_clip=error_clip,
                 )
             else:
-                if _in_legacy_dygraph():
-                    var = ParamBase(
-                        d.shape(),
-                        d.dtype(),
-                        type=orig_var_type,
-                        name=new_name,
-                        stop_gradient=stop_gradient,
-                        trainable=trainable,
-                        optimize_attr=optimize_attr,
-                        regularizer=regularizer,
-                        error_clip=error_clip,
-                    )
-                else:
                     var = Parameter(
                         self,
                         d.shape(),
@@ -3946,9 +3927,6 @@ def create_parameter(self, *args, **kwargs):
         if in_dygraph_mode():
             param = EagerParamBase(*args, **kwargs)
         else:
-            if _in_legacy_dygraph():
-                param = ParamBase(*args, **kwargs)
-            else:
                 param = Parameter(global_block, *args, **kwargs)
 
         if 'initializer' in kwargs:
@@ -4262,20 +4240,6 @@ def _copy_param_info_from(self, other):
                     name=v.name,
                 )
             else:
-                if _in_legacy_dygraph():
-                    new_p = ParamBase(
-                        shape=v.shape,
-                        dtype=v.dtype,
-                        type=v.type,
-                        lod_level=v.lod_level,
-                        stop_gradient=p.stop_gradient,
-                        trainable=p.trainable,
-                        optimize_attr=p.optimize_attr,
-                        regularizer=p.regularizer,
-                        error_clip=p.error_clip,
-                        name=v.name,
-                    )
-                else:
                     new_p = Parameter(
                         block=self,
                         shape=v.shape,

python/paddle/fluid/layers/layer_function_generator.py

@@ -272,7 +272,6 @@ def func(x, name=None):
             op = getattr(_C_ops, op_type)
             return op(x)
         # TODO(dev): Because some ops' yaml has not been migrated.
-        # Replace it with _in_legacy_dygraph while all yaml work is done.
         if in_dygraph_mode() and hasattr(_legacy_C_ops, op_type):
             op = getattr(_legacy_C_ops, op_type)
             return op(x)

python/paddle/fluid/tests/unittests/op_test.py

@@ -38,7 +38,6 @@
     _dygraph_tracer,
     _enable_legacy_dygraph,
     _in_eager_without_dygraph_check,
-    _in_legacy_dygraph,
     _test_eager_guard,
 )
 from paddle.fluid.op import Operator
@@ -716,7 +715,7 @@ def create_var(np_value, name, is_input, if_return_inputs_grad_dict):
 
                 if if_return_inputs_grad_dict:
                     v.stop_gradient = False
-                    if not _in_legacy_dygraph():
+                    if hasattr(v, "retain_grads"):
                         v.retain_grads()
 
                 if has_lod:
@@ -2515,22 +2514,14 @@ def _get_dygraph_grad(
                 for no_grad_val in no_grad_set:
                     del inputs[no_grad_val]
 
-                if not _in_legacy_dygraph():
-                    core.eager.run_backward(
+                core.eager.run_backward(
                         fluid.layers.utils.flatten(outputs), grad_outputs, False
                     )
-                    grad_inputs = []
-                    for inputs_list in inputs.values():
+                grad_inputs = []
+                for inputs_list in inputs.values():
                         for inp in inputs_list:
                             grad_inputs.append(inp.grad.numpy())
-                    return grad_inputs
-                else:
-                    grad_inputs = paddle.grad(
-                        outputs=fluid.layers.utils.flatten(outputs),
-                        inputs=fluid.layers.utils.flatten(inputs),
-                        grad_outputs=grad_outputs,
-                    )
-                    return [grad.numpy() for grad in grad_inputs]
+                return grad_inputs
 
     @staticmethod
     def _numpy_to_lod_tensor(np_value, lod, place):

python/paddle/framework/__init__.py

@@ -64,7 +64,6 @@
 
 from ..fluid.layer_helper import LayerHelper  # noqa: F401
 from ..fluid.framework import in_dygraph_mode  # noqa: F401
-from ..fluid.framework import _in_legacy_dygraph  # noqa: F401
 from ..fluid.framework import _global_flags  # noqa: F401
 from ..fluid.framework import _apply_pass  # noqa: F401
 from ..fluid.framework import switch_main_program

python/paddle/nn/functional/norm.py

@@ -17,8 +17,8 @@
 # TODO: define normalization api
 import paddle
 import paddle.fluid as fluid
-from paddle import _C_ops, _legacy_C_ops, in_dynamic_mode
-from paddle.fluid.framework import _in_legacy_dygraph, in_dygraph_mode
+from paddle import _C_ops, in_dynamic_mode
+from paddle.fluid.framework import in_dygraph_mode
 
 from ...fluid import dygraph_utils
 from ...fluid.data_feeder import check_type, check_variable_and_dtype
@@ -336,54 +336,43 @@ def layer_norm(
         out, _, _ = _C_ops.layer_norm(x, weight, bias, epsilon, begin_norm_axis)
         return out
 
-    if _in_legacy_dygraph():
-        out, _, _ = _legacy_C_ops.layer_norm(
-            x,
-            weight,
-            bias,
-            'epsilon',
-            epsilon,
-            'begin_norm_axis',
-            begin_norm_axis,
+    else:
+        check_variable_and_dtype(
+            x, 'input', ['float16', 'float32', 'float64'], 'LayerNorm'
+        )
+
+        inputs = dict()
+        inputs['X'] = [x]
+        if weight:
+            inputs['Scale'] = [weight]
+        if bias:
+            inputs['Bias'] = [bias]
+        attrs = {"epsilon": epsilon, "begin_norm_axis": begin_norm_axis}
+
+        # create output
+        helper = LayerHelper('layer_norm', **locals())
+
+        dtype = x.dtype
+        mean_out = helper.create_variable_for_type_inference(
+            dtype=dtype, stop_gradient=True
+        )
+        variance_out = helper.create_variable_for_type_inference(
+            dtype=dtype, stop_gradient=True
+        )
+        layer_norm_out = helper.create_variable_for_type_inference(dtype)
+
+        helper.append_op(
+            type="layer_norm",
+            inputs=inputs,
+            outputs={
+                "Y": layer_norm_out,
+                "Mean": mean_out,
+                "Variance": variance_out,
+            },
+            attrs={"epsilon": epsilon, "begin_norm_axis": begin_norm_axis},
         )
-        return out
 
-    check_variable_and_dtype(
-        x, 'input', ['float16', 'float32', 'float64'], 'LayerNorm'
-    )
-
-    inputs = dict()
-    inputs['X'] = [x]
-    if weight:
-        inputs['Scale'] = [weight]
-    if bias:
-        inputs['Bias'] = [bias]
-    attrs = {"epsilon": epsilon, "begin_norm_axis": begin_norm_axis}
-
-    # create output
-    helper = LayerHelper('layer_norm', **locals())
-
-    dtype = x.dtype
-    mean_out = helper.create_variable_for_type_inference(
-        dtype=dtype, stop_gradient=True
-    )
-    variance_out = helper.create_variable_for_type_inference(
-        dtype=dtype, stop_gradient=True
-    )
-    layer_norm_out = helper.create_variable_for_type_inference(dtype)
-
-    helper.append_op(
-        type="layer_norm",
-        inputs=inputs,
-        outputs={
-            "Y": layer_norm_out,
-            "Mean": mean_out,
-            "Variance": variance_out,
-        },
-        attrs={"epsilon": epsilon, "begin_norm_axis": begin_norm_axis},
-    )
-
-    return helper.append_activation(layer_norm_out)
+        return helper.append_activation(layer_norm_out)
 
 
 def instance_norm(