diff --git a/paddle/fluid/API.spec b/paddle/fluid/API.spec
index 0a0c3663f3c2f5fd7a638d7e5b1ee130d20e234b..f435a79871ece81fa744df2c3b894f4fc997c900 100644
--- a/paddle/fluid/API.spec
+++ b/paddle/fluid/API.spec
@@ -24,6 +24,7 @@ paddle.fluid.Variable.clear_gradient (ArgSpec(args=['self'], varargs=None, keywo
 paddle.fluid.Variable.detach (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '0730b2d310b014d9b0a903b2034757d7'))
 paddle.fluid.Variable.gradient (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '86b246bfaf20f3058e91927abbcf9fb9'))
 paddle.fluid.Variable.numpy (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '7536e8feb56d827875943e7f01d406fc'))
+paddle.fluid.Variable.set_value (ArgSpec(args=['self', 'value'], varargs=None, keywords=None, defaults=None), ('document', 'c424b9e763ff51c38a6917f98026fe7d'))
 paddle.fluid.Variable.to_string (ArgSpec(args=['self', 'throw_on_error', 'with_details'], varargs=None, keywords=None, defaults=(False,)), ('document', '31f359a2c074f26dc0ffff296fc3983f'))
 paddle.fluid.Executor ('paddle.fluid.executor.Executor', ('document', '34e8c1769313fbeff7817212dda6259e'))
 paddle.fluid.Executor.__init__ (ArgSpec(args=['self', 'place'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
@@ -575,7 +576,7 @@ paddle.fluid.dygraph.Layer.clear_gradients (ArgSpec(args=['self'], varargs=None,
 paddle.fluid.dygraph.Layer.create_parameter (ArgSpec(args=['self', 'attr', 'shape', 'dtype', 'is_bias', 'default_initializer'], varargs=None, keywords=None, defaults=(False, None)), ('document', 'a6420ca1455366eaaf972191612de0b6'))
 paddle.fluid.dygraph.Layer.create_variable (ArgSpec(args=['self', 'name', 'persistable', 'dtype', 'type'], varargs=None, keywords=None, defaults=(None, None, None, VarType.LOD_TENSOR)), ('document', '171cccfceba636d5bbf7bbae672945d8'))
 paddle.fluid.dygraph.Layer.eval (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Layer.forward (ArgSpec(args=['self'], varargs='inputs', keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
+paddle.fluid.dygraph.Layer.forward (ArgSpec(args=['self'], varargs='inputs', keywords='kwargs', defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Layer.full_name (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '23ce4f961f48ed0f79cadf93a3938ed2'))
 paddle.fluid.dygraph.Layer.load_dict (ArgSpec(args=['self', 'stat_dict', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Layer.parameters (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '5aec25a854eb57abc798dccccbb507d5'))
@@ -585,7 +586,7 @@ paddle.fluid.dygraph.Layer.train (ArgSpec(args=['self'], varargs=None, keywords=
 paddle.fluid.dygraph.__impl__ (ArgSpec(args=['func'], varargs=None, keywords=None, defaults=()), ('document', '75d1d3afccc8b39cdebf05cb1f5969f9'))
 paddle.fluid.dygraph.guard (ArgSpec(args=['place'], varargs=None, keywords=None, defaults=(None,)), ('document', '7071320ffe2eec9aacdae574951278c6'))
 paddle.fluid.dygraph.to_variable (ArgSpec(args=['value', 'block', 'name'], varargs=None, keywords=None, defaults=(None, None)), ('document', '0e69fa3666f15dd01b6e3e270b9371cd'))
-paddle.fluid.dygraph.Conv2D ('paddle.fluid.dygraph.nn.Conv2D', ('document', 'baafe7ae0d3a61ae79cf4c7443e2c37c'))
+paddle.fluid.dygraph.Conv2D ('paddle.fluid.dygraph.nn.Conv2D', ('document', '0b6acb9cc7fbb4f5b129e1f6dd985581'))
 paddle.fluid.dygraph.Conv2D.__init__ (ArgSpec(args=['self', 'name_scope', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'dtype'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None, 'float32')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv2D.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Conv2D.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -601,7 +602,7 @@ paddle.fluid.dygraph.Conv2D.parameters (ArgSpec(args=['self', 'include_sublayers
 paddle.fluid.dygraph.Conv2D.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv2D.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Conv2D.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Conv3D ('paddle.fluid.dygraph.nn.Conv3D', ('document', '8b756aaca65af9594cc574d9a5d2b055'))
+paddle.fluid.dygraph.Conv3D ('paddle.fluid.dygraph.nn.Conv3D', ('document', '50412bd3fbf3557a8ef48e25c6517025'))
 paddle.fluid.dygraph.Conv3D.__init__ (ArgSpec(args=['self', 'name_scope', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv3D.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Conv3D.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -617,7 +618,7 @@ paddle.fluid.dygraph.Conv3D.parameters (ArgSpec(args=['self', 'include_sublayers
 paddle.fluid.dygraph.Conv3D.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv3D.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Conv3D.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Pool2D ('paddle.fluid.dygraph.nn.Pool2D', ('document', 'e9331666e47a38586c8809a23cbaf7de'))
+paddle.fluid.dygraph.Pool2D ('paddle.fluid.dygraph.nn.Pool2D', ('document', '50e6fd200e42859daf2924ecb0561ada'))
 paddle.fluid.dygraph.Pool2D.__init__ (ArgSpec(args=['self', 'name_scope', 'pool_size', 'pool_type', 'pool_stride', 'pool_padding', 'global_pooling', 'use_cudnn', 'ceil_mode', 'exclusive', 'dtype'], varargs=None, keywords=None, defaults=(-1, 'max', 1, 0, False, True, False, True, VarType.FP32)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Pool2D.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Pool2D.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -633,7 +634,7 @@ paddle.fluid.dygraph.Pool2D.parameters (ArgSpec(args=['self', 'include_sublayers
 paddle.fluid.dygraph.Pool2D.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Pool2D.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Pool2D.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.FC ('paddle.fluid.dygraph.nn.FC', ('document', '1d64242f03f2aca2307e94590b552430'))
+paddle.fluid.dygraph.FC ('paddle.fluid.dygraph.nn.FC', ('document', '2f73ae00e57c67454c6aa7e911d9bfd6'))
 paddle.fluid.dygraph.FC.__init__ (ArgSpec(args=['self', 'name_scope', 'size', 'num_flatten_dims', 'param_attr', 'bias_attr', 'act', 'is_test', 'dtype'], varargs=None, keywords=None, defaults=(1, None, None, None, False, 'float32')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.FC.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.FC.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -649,7 +650,7 @@ paddle.fluid.dygraph.FC.parameters (ArgSpec(args=['self', 'include_sublayers'],
 paddle.fluid.dygraph.FC.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.FC.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.FC.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.BatchNorm ('paddle.fluid.dygraph.nn.BatchNorm', ('document', '0b609e10e4d417c91d346f887d16771c'))
+paddle.fluid.dygraph.BatchNorm ('paddle.fluid.dygraph.nn.BatchNorm', ('document', '390fb9b986423ec6680731ffc7cf24ab'))
 paddle.fluid.dygraph.BatchNorm.__init__ (ArgSpec(args=['self', 'name_scope', 'num_channels', 'act', 'is_test', 'momentum', 'epsilon', 'param_attr', 'bias_attr', 'dtype', 'data_layout', 'in_place', 'moving_mean_name', 'moving_variance_name', 'do_model_average_for_mean_and_var', 'fuse_with_relu', 'use_global_stats', 'trainable_statistics'], varargs=None, keywords=None, defaults=(None, False, 0.9, 1e-05, None, None, 'float32', 'NCHW', False, None, None, False, False, False, False)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.BatchNorm.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.BatchNorm.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -665,7 +666,7 @@ paddle.fluid.dygraph.BatchNorm.parameters (ArgSpec(args=['self', 'include_sublay
 paddle.fluid.dygraph.BatchNorm.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.BatchNorm.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.BatchNorm.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Embedding ('paddle.fluid.dygraph.nn.Embedding', ('document', 'ecf8dc4149f005cd30eddc0dd343454f'))
+paddle.fluid.dygraph.Embedding ('paddle.fluid.dygraph.nn.Embedding', ('document', 'b1b1ed9dc2125c3e16ee08113605fcb4'))
 paddle.fluid.dygraph.Embedding.__init__ (ArgSpec(args=['self', 'name_scope', 'size', 'is_sparse', 'is_distributed', 'padding_idx', 'param_attr', 'dtype'], varargs=None, keywords=None, defaults=(False, False, None, None, 'float32')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Embedding.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Embedding.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -681,7 +682,7 @@ paddle.fluid.dygraph.Embedding.parameters (ArgSpec(args=['self', 'include_sublay
 paddle.fluid.dygraph.Embedding.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Embedding.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Embedding.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.GRUUnit ('paddle.fluid.dygraph.nn.GRUUnit', ('document', '5308e42b6a6db4681ce5ee9e94983986'))
+paddle.fluid.dygraph.GRUUnit ('paddle.fluid.dygraph.nn.GRUUnit', ('document', '389e860e455b67aab1f4d472ac9d7e49'))
 paddle.fluid.dygraph.GRUUnit.__init__ (ArgSpec(args=['self', 'name_scope', 'size', 'param_attr', 'bias_attr', 'activation', 'gate_activation', 'origin_mode', 'dtype'], varargs=None, keywords=None, defaults=(None, None, 'tanh', 'sigmoid', False, 'float32')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.GRUUnit.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.GRUUnit.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -697,7 +698,7 @@ paddle.fluid.dygraph.GRUUnit.parameters (ArgSpec(args=['self', 'include_sublayer
 paddle.fluid.dygraph.GRUUnit.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.GRUUnit.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.GRUUnit.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.LayerNorm ('paddle.fluid.dygraph.nn.LayerNorm', ('document', 'b44f5d3d10386c460094e21f24ff272b'))
+paddle.fluid.dygraph.LayerNorm ('paddle.fluid.dygraph.nn.LayerNorm', ('document', '8bc39f59fe2d3713bc143fdf1222a63b'))
 paddle.fluid.dygraph.LayerNorm.__init__ (ArgSpec(args=['self', 'name_scope', 'scale', 'shift', 'begin_norm_axis', 'epsilon', 'param_attr', 'bias_attr', 'act'], varargs=None, keywords=None, defaults=(True, True, 1, 1e-05, None, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.LayerNorm.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.LayerNorm.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -713,7 +714,7 @@ paddle.fluid.dygraph.LayerNorm.parameters (ArgSpec(args=['self', 'include_sublay
 paddle.fluid.dygraph.LayerNorm.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.LayerNorm.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.LayerNorm.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.NCE ('paddle.fluid.dygraph.nn.NCE', ('document', '2d579e8d9ce31bb29e079e5f6108fc73'))
+paddle.fluid.dygraph.NCE ('paddle.fluid.dygraph.nn.NCE', ('document', '993aeea9be436e9c709a758795cb23e9'))
 paddle.fluid.dygraph.NCE.__init__ (ArgSpec(args=['self', 'name_scope', 'num_total_classes', 'sample_weight', 'param_attr', 'bias_attr', 'num_neg_samples', 'sampler', 'custom_dist', 'seed', 'is_sparse'], varargs=None, keywords=None, defaults=(None, None, None, None, 'uniform', None, 0, False)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.NCE.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.NCE.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -729,7 +730,7 @@ paddle.fluid.dygraph.NCE.parameters (ArgSpec(args=['self', 'include_sublayers'],
 paddle.fluid.dygraph.NCE.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.NCE.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.NCE.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.PRelu ('paddle.fluid.dygraph.nn.PRelu', ('document', 'd395ed163b4cf86e7207968f27bc1c11'))
+paddle.fluid.dygraph.PRelu ('paddle.fluid.dygraph.nn.PRelu', ('document', 'da956af1676b08bf15553751a3643b55'))
 paddle.fluid.dygraph.PRelu.__init__ (ArgSpec(args=['self', 'name_scope', 'mode', 'param_attr'], varargs=None, keywords=None, defaults=(None,)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.PRelu.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.PRelu.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -745,7 +746,7 @@ paddle.fluid.dygraph.PRelu.parameters (ArgSpec(args=['self', 'include_sublayers'
 paddle.fluid.dygraph.PRelu.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.PRelu.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.PRelu.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.BilinearTensorProduct ('paddle.fluid.dygraph.nn.BilinearTensorProduct', ('document', '310140d784933928a27db9a7af4761e8'))
+paddle.fluid.dygraph.BilinearTensorProduct ('paddle.fluid.dygraph.nn.BilinearTensorProduct', ('document', 'be70d0f6d43729d9cb80c9a34ed5f26b'))
 paddle.fluid.dygraph.BilinearTensorProduct.__init__ (ArgSpec(args=['self', 'name_scope', 'size', 'name', 'act', 'param_attr', 'bias_attr'], varargs=None, keywords=None, defaults=(None, None, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.BilinearTensorProduct.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.BilinearTensorProduct.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -761,7 +762,7 @@ paddle.fluid.dygraph.BilinearTensorProduct.parameters (ArgSpec(args=['self', 'in
 paddle.fluid.dygraph.BilinearTensorProduct.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.BilinearTensorProduct.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.BilinearTensorProduct.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Conv2DTranspose ('paddle.fluid.dygraph.nn.Conv2DTranspose', ('document', '918fa8ad8a62ff424c842fb8a840bf7a'))
+paddle.fluid.dygraph.Conv2DTranspose ('paddle.fluid.dygraph.nn.Conv2DTranspose', ('document', 'cf23c905abc00b07603dfa71a432d6f7'))
 paddle.fluid.dygraph.Conv2DTranspose.__init__ (ArgSpec(args=['self', 'name_scope', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv2DTranspose.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Conv2DTranspose.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -777,7 +778,7 @@ paddle.fluid.dygraph.Conv2DTranspose.parameters (ArgSpec(args=['self', 'include_
 paddle.fluid.dygraph.Conv2DTranspose.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv2DTranspose.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Conv2DTranspose.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.Conv3DTranspose ('paddle.fluid.dygraph.nn.Conv3DTranspose', ('document', 'cd99906d0813609ddea3fb6a2ac900dc'))
+paddle.fluid.dygraph.Conv3DTranspose ('paddle.fluid.dygraph.nn.Conv3DTranspose', ('document', '91ba132bc690eaf76eabdbde8f87e4a0'))
 paddle.fluid.dygraph.Conv3DTranspose.__init__ (ArgSpec(args=['self', 'name_scope', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv3DTranspose.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.Conv3DTranspose.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -793,7 +794,7 @@ paddle.fluid.dygraph.Conv3DTranspose.parameters (ArgSpec(args=['self', 'include_
 paddle.fluid.dygraph.Conv3DTranspose.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.Conv3DTranspose.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.Conv3DTranspose.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.GroupNorm ('paddle.fluid.dygraph.nn.GroupNorm', ('document', '4d65fc6b00970e3b5c5dd0abeacd47cb'))
+paddle.fluid.dygraph.GroupNorm ('paddle.fluid.dygraph.nn.GroupNorm', ('document', '72c125b07bdd1e612607dc77039b2722'))
 paddle.fluid.dygraph.GroupNorm.__init__ (ArgSpec(args=['self', 'name_scope', 'groups', 'epsilon', 'param_attr', 'bias_attr', 'act', 'data_layout'], varargs=None, keywords=None, defaults=(1e-05, None, None, None, 'NCHW')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.GroupNorm.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.GroupNorm.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -809,7 +810,7 @@ paddle.fluid.dygraph.GroupNorm.parameters (ArgSpec(args=['self', 'include_sublay
 paddle.fluid.dygraph.GroupNorm.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.GroupNorm.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.GroupNorm.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.SpectralNorm ('paddle.fluid.dygraph.nn.SpectralNorm', ('document', 'f400a29393aa95fff829b4a6111e2952'))
+paddle.fluid.dygraph.SpectralNorm ('paddle.fluid.dygraph.nn.SpectralNorm', ('document', '8f5cfbc431a8b4b44b605cde8b0381ef'))
 paddle.fluid.dygraph.SpectralNorm.__init__ (ArgSpec(args=['self', 'name_scope', 'dim', 'power_iters', 'eps', 'name'], varargs=None, keywords=None, defaults=(0, 1, 1e-12, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.SpectralNorm.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.SpectralNorm.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
@@ -825,7 +826,7 @@ paddle.fluid.dygraph.SpectralNorm.parameters (ArgSpec(args=['self', 'include_sub
 paddle.fluid.dygraph.SpectralNorm.state_dict (ArgSpec(args=['self', 'destination', 'include_sublayers'], varargs=None, keywords=None, defaults=(None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.SpectralNorm.sublayers (ArgSpec(args=['self', 'include_sublayers'], varargs=None, keywords=None, defaults=(True,)), ('document', '00a881005ecbc96578faf94513bf0d62'))
 paddle.fluid.dygraph.SpectralNorm.train (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
-paddle.fluid.dygraph.TreeConv ('paddle.fluid.dygraph.nn.TreeConv', ('document', '1e3104dea2482f6b79cf7a7ac9a343ab'))
+paddle.fluid.dygraph.TreeConv ('paddle.fluid.dygraph.nn.TreeConv', ('document', '6e175a7bf2a43ae6c0f3a8a54bd69afe'))
 paddle.fluid.dygraph.TreeConv.__init__ (ArgSpec(args=['self', 'name_scope', 'output_size', 'num_filters', 'max_depth', 'act', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(1, 2, 'tanh', None, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.dygraph.TreeConv.add_parameter (ArgSpec(args=['self', 'name', 'parameter'], varargs=None, keywords=None, defaults=None), ('document', 'f35ab374c7d5165c3daf3bd64a5a2ec1'))
 paddle.fluid.dygraph.TreeConv.add_sublayer (ArgSpec(args=['self', 'name', 'sublayer'], varargs=None, keywords=None, defaults=None), ('document', '839ff3c0534677ba6ad8735c3fd4e995'))
diff --git a/python/paddle/fluid/dygraph/layers.py b/python/paddle/fluid/dygraph/layers.py
index afb18ed505bd6f31c4aa2a0ad4feafdaa1da28f1..e53f0dd2b669ccb095ba2b25e18354c53935fe1d 100644
--- a/python/paddle/fluid/dygraph/layers.py
+++ b/python/paddle/fluid/dygraph/layers.py
@@ -150,20 +150,20 @@ class Layer(core.Layer):
             if p.trainable:
                 p.clear_gradient()
 
-    def _build_once(self, *args):
+    def _build_once(self, *args, **kwargs):
         pass
 
-    def __call__(self, *inputs):
+    def __call__(self, *inputs, **kwargs):
         if not self._built:
-            self._build_once(*inputs)
+            self._build_once(*inputs, **kwargs)
             if parallel_helper._is_data_parallel_mode():
                 parallel_helper._broadcast_parameters(self._parameters.values())
 
-        outputs = self.forward(*inputs)
+        outputs = self.forward(*inputs, **kwargs)
         self._built = True
         return outputs
 
-    def forward(self, *inputs):
+    def forward(self, *inputs, **kwargs):
         raise NotImplementedError
 
     def backward(self, *inputs):
@@ -216,6 +216,8 @@ class Layer(core.Layer):
             return object.__getattribute__(self, name)
 
     def __setattr__(self, name, value):
+        if isinstance(getattr(type(self), name, None), property):
+            object.__setattr__(self, name, value)
         if isinstance(value, framework.Parameter):
             params = self.__dict__.get('_parameters', None)
             if params is None:
@@ -226,6 +228,11 @@ class Layer(core.Layer):
                 tensor = var.get_tensor()
                 tensor.set(self._loaddict_holder[value.name].numpy(),
                            framework._current_expected_place())
+            if name in params:
+                # remove unused param in tracer
+                if framework._dygraph_tracer_ is not None:
+                    framework._dygraph_tracer_._vars.pop(params[name].name,
+                                                         None)
             params[name] = value
         elif isinstance(value, core.Layer):
             layers = self.__dict__.get('_sub_layers', None)
diff --git a/python/paddle/fluid/dygraph/nn.py b/python/paddle/fluid/dygraph/nn.py
index 27aeda45d80657de2a0d43c55bb23083c264f1fc..7802dd3cb2f7101d6f6a435406346dcafe182f4d 100644
--- a/python/paddle/fluid/dygraph/nn.py
+++ b/python/paddle/fluid/dygraph/nn.py
@@ -83,7 +83,7 @@ class Conv2D(layers.Layer):
             H_{out}&= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]} + 1 \\\\
             W_{out}&= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]} + 1
 
-    Args:
+    Parameters:
         name_scope(str) : The name for this class.
         num_filters(int): The number of filter. It is as same as the output
             image channel.
@@ -119,6 +119,10 @@ class Conv2D(layers.Layer):
         act (str): Activation type, if it is set to None, activation is not appended.
             Default: None
 
+    Attributes:
+        weight (Parameter): the learnable weights of filter of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Raises:
         ValueError: If the shapes of input, filter_size, stride, padding and
                     groups mismatch.
@@ -205,6 +209,22 @@ class Conv2D(layers.Layer):
             dtype=self._dtype,
             is_bias=True)
 
+    @property
+    def weight(self):
+        return self._filter_param
+
+    @weight.setter
+    def weight(self, value):
+        self._filter_param = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, input):
         pre_bias = self._helper.create_variable_for_type_inference(
             dtype=self._dtype)
@@ -288,7 +308,7 @@ class Conv3D(layers.Layer):
             H_{out}&= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{strides[1]} + 1 \\\\
             W_{out}&= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{strides[2]} + 1
 
-    Args:
+    Parameters:
         name_scope(str) : The name for this class.
         num_filters(int): The number of filter. It is as same as the output image channel.
         filter_size (int|tuple|None): The filter size. If filter_size is a tuple,
@@ -323,6 +343,10 @@ class Conv3D(layers.Layer):
         act (str): Activation type, if it is set to None, activation is not appended.
             Default: None.
 
+    Attributes:
+        weight (Parameter): the learnable weights of filters of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
         Variable: The tensor variable storing the convolution and \
                   non-linearity activation result.
@@ -405,6 +429,22 @@ class Conv3D(layers.Layer):
             dtype=self._dtype,
             is_bias=True)
 
+    @property
+    def weight(self):
+        return self._filter_param
+
+    @weight.setter
+    def weight(self, value):
+        self._filter_param = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, input):
         pre_bias = self._helper.create_variable_for_type_inference(
             dtype=self._dtype)
@@ -425,15 +465,17 @@ class Conv3D(layers.Layer):
                 'use_mkldnn': False
             })
 
-        pre_act = self._helper.create_variable_for_type_inference(
-            dtype=self._dtype)
-
-        self._helper.append_op(
-            type='elementwise_add',
-            inputs={'X': [pre_bias],
-                    'Y': [self._bias_param]},
-            outputs={'Out': [pre_act]},
-            attrs={'axis': 1})
+        if self._bias_param is not None:
+            pre_act = self._helper.create_variable_for_type_inference(
+                dtype=self._dtype)
+            self._helper.append_op(
+                type='elementwise_add',
+                inputs={'X': [pre_bias],
+                        'Y': [self._bias_param]},
+                outputs={'Out': [pre_act]},
+                attrs={'axis': 1})
+        else:
+            pre_act = pre_bias
 
         return self._helper.append_activation(pre_act, act=self._act)
 
@@ -489,7 +531,7 @@ class Conv3DTranspose(layers.Layer):
            H_{out} &= (H_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (H_f - 1) + 1 \\\\
            W_{out} &= (W_{in} - 1) * strides[2] - 2 * paddings[2] + dilations[2] * (W_f - 1) + 1
 
-    Args:
+    Parameters:
         name_scope(str) : The name for this class.
         num_filters(int): The number of the filter. It is as same as the output
             image channel.
@@ -531,6 +573,10 @@ class Conv3DTranspose(layers.Layer):
         name(str|None): A name for this layer(optional). If set None, the layer
             will be named automatically.
 
+    Attributes:
+        weight (Parameter): the learnable weights of filters of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
         Variable: The tensor variable storing the convolution transpose result.
 
@@ -627,6 +673,22 @@ class Conv3DTranspose(layers.Layer):
                 dtype=self._dtype,
                 is_bias=True)
 
+    @property
+    def weight(self):
+        return self._img_filter
+
+    @weight.setter
+    def weight(self, value):
+        self._img_filter = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, input):
         pre_bias = self._helper.create_variable_for_type_inference(
             dtype=self._dtype)
@@ -667,7 +729,7 @@ class Pool2D(layers.Layer):
     Parameters(ksize, strides, paddings) are two elements. These two elements represent height and width, respectively.
     The input(X) size and output(Out) size may be different.
 
-    Args:
+    Parameters:
         name_scope(str) : The name of this class.
         pool_size (int|list|tuple): The pool kernel size. If pool kernel size is a tuple or list,
             it must contain two integers, (pool_size_Height, pool_size_Width).
@@ -824,7 +886,7 @@ class FC(layers.Layer):
             out.data = [[0.18669507, 0.1893476]]
             out.shape = (1, 2)
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         size(int): The number of output units in this layer.
         num_flatten_dims (int): The fc layer can accept an input tensor with more than
@@ -845,6 +907,10 @@ class FC(layers.Layer):
         is_test(bool): A flag indicating whether execution is in test phase. Default: False
         dtype(str): Dtype used for weight
 
+    Attributes:
+        weight (list of Parameter): the learnable weights of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Raises:
         ValueError: If rank of the input tensor is less than 2.
 
@@ -883,15 +949,6 @@ class FC(layers.Layer):
         self._act = act
         self.__w = list()
 
-    @property
-    def _w(self, i=0):
-        return self.__w[i]
-
-    @_w.setter
-    def _w(self, value, i=0):
-        assert isinstance(value, Parameter)
-        self.__w[i] = value
-
     def _build_once(self, input):
         i = 0
         for inp, param in self._helper.iter_inputs_and_params(input,
@@ -916,6 +973,36 @@ class FC(layers.Layer):
         self._b = self.create_parameter(
             attr=self._bias_attr, shape=size, dtype=self._dtype, is_bias=True)
 
+    # TODO(songyouwei): We should remove _w property
+    @property
+    def _w(self, i=0):
+        return self.__w[i]
+
+    @_w.setter
+    def _w(self, value, i=0):
+        assert isinstance(self.__w[i], Variable)
+        self.__w[i].set_value(value)
+
+    @property
+    def weight(self):
+        if len(self.__w) > 1:
+            return self.__w
+        else:
+            return self.__w[0]
+
+    @weight.setter
+    def weight(self, value):
+        if len(self.__w) == 1:
+            self.__w[0] = value
+
+    @property
+    def bias(self):
+        return self._b
+
+    @bias.setter
+    def bias(self, value):
+        self._b = value
+
     def forward(self, input):
         mul_results = list()
         i = 0
@@ -1000,7 +1087,7 @@ class BatchNorm(layers.Layer):
         \\sigma_{\\beta}^{2} + \\epsilon}}  \\\\
         y_i &\\gets \\gamma \\hat{x_i} + \\beta
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         act(str|None): Activation type, linear|relu|prelu|...
         is_test (bool): A flag indicating whether it is in
@@ -1184,7 +1271,7 @@ class Embedding(layers.Layer):
     :attr:`input`.
     All the input variables are passed in as local variables to the LayerHelper constructor
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         size(tuple|list): The shape of the look up table parameter. It should have two elements which indicate the size
             of the dictionary of embeddings and the size of each embedding vector respectively.
@@ -1196,6 +1283,9 @@ class Embedding(layers.Layer):
         param_attr(ParamAttr): Parameters for this layer. Default: None.
         dtype(np.dtype|core.VarDesc.VarType|str): The type of data : float32, float_16, int etc. Default: 'float32'.
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+
     Returns:
         Variable: The tensor variable storing the embeddings of the \
                   supplied inputs.
@@ -1246,6 +1336,14 @@ class Embedding(layers.Layer):
             dtype=self._dtype,
             is_bias=False)
 
+    @property
+    def weight(self):
+        return self._w
+
+    @weight.setter
+    def weight(self, value):
+        self._w = value
+
     def forward(self, input):
         out = self._helper.create_variable_for_type_inference(self._dtype)
         self._helper.append_op(
@@ -1291,7 +1389,7 @@ class LayerNorm(layers.Layer):
 
     * :math:`b`: the trainable bias parameter.
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         scale(bool): Whether to learn the adaptive gain :math:`g` after
             normalization. Default: True.
@@ -1451,7 +1549,7 @@ class GRUUnit(layers.Layer):
     This layer has three outputs :math:`h_t`, :math:`dot(r_t, h_{t-1})`
     and concatenation of :math:`u_t`, :math:`r_t` and :math:`m_t`.
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         size (int): The input dimension value.
         param_attr(ParamAttr|None): The parameter attribute for the learnable
@@ -1481,6 +1579,10 @@ class GRUUnit(layers.Layer):
                                   Default: 'sigmoid'
         dtype(str): The dtype of the layers. Default: 'float32'
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+        bias (Parameter): the learnable bias of this layer.
+
     Returns:
         tuple: The hidden value, reset-hidden value and gate values.
 
@@ -1515,6 +1617,7 @@ class GRUUnit(layers.Layer):
                  origin_mode=False,
                  dtype='float32'):
         super(GRUUnit, self).__init__(name_scope, dtype)
+        self._bias_attr = bias_attr
 
         activation_dict = dict(
             identity=0,
@@ -1532,9 +1635,26 @@ class GRUUnit(layers.Layer):
 
         # create bias
         bias_size = [1, 3 * size]
+        self._bias_size = bias_size
         self._bias = self.create_parameter(
             attr=bias_attr, shape=bias_size, dtype=dtype, is_bias=True)
 
+    @property
+    def weight(self):
+        return self._weight
+
+    @weight.setter
+    def weight(self, value):
+        self._weight = value
+
+    @property
+    def bias(self):
+        return self._bias
+
+    @bias.setter
+    def bias(self, value):
+        self._bias = value
+
     def forward(self, input, hidden):
         inputs = {'Input': input, 'HiddenPrev': hidden, 'Weight': self._weight}
         if self._bias:
@@ -1567,7 +1687,7 @@ class NCE(layers.Layer):
     `Noise-contrastive estimation: A new estimation principle for unnormalized statistical models <http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf>`_ .
     By default this operator uses a uniform distribution for sampling.
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         num_total_classes (int): Total number of classes in all samples
         param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
@@ -1590,6 +1710,10 @@ class NCE(layers.Layer):
         seed (int): The seed used in sampler. Default: 0.
         is_sparse(bool): The flag indicating whether to use sparse update, the weight@GRAD and bias@GRAD will be changed to SelectedRows. Default: False.
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
         Variable: The output nce loss.
 
@@ -1763,6 +1887,22 @@ class NCE(layers.Layer):
             self._inputs['Bias'] = self._b
         self._inputs['Weight'] = self._w
 
+    @property
+    def weight(self):
+        return self._w
+
+    @weight.setter
+    def weight(self, value):
+        self._w = value
+
+    @property
+    def bias(self):
+        return self._b
+
+    @bias.setter
+    def bias(self, value):
+        self._b = value
+
     def forward(self, input, label, sample_weight=None):
         assert isinstance(input, Variable)
         assert isinstance(label, Variable)
@@ -1797,7 +1937,7 @@ class PRelu(layers.Layer):
     .. math::
         y = \max(0, x) + \\alpha * \min(0, x)
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         mode (str): The mode for weight sharing. It supports all, channel
           and element. all: all elements share same weight
@@ -1806,6 +1946,9 @@ class PRelu(layers.Layer):
         param_attr(ParamAttr|None): The parameter attribute for the learnable
           weight (alpha).
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+
     Returns:
         Variable: The output tensor with the same shape as input.
 
@@ -1849,6 +1992,14 @@ class PRelu(layers.Layer):
             is_bias=False,
             default_initializer=Constant(1.0))
 
+    @property
+    def weight(self):
+        return self._alpha
+
+    @weight.setter
+    def weight(self, value):
+        self._alpha = value
+
     def forward(self, input):
 
         out = self._helper.create_variable_for_type_inference(self._dtype)
@@ -1878,7 +2029,7 @@ class BilinearTensorProduct(layers.Layer):
      - :math:`out_{i}`: the i-th element of out, shape is [batch_size, size].
      - :math:`y^\mathrm{T}`: the transpose of :math:`y_{2}`.
 
-    Args:
+    Parameters:
        name_scope(str): The name of this class.
        size (int): The dimension of this layer.
        act (str): Activation to be applied to the output of this layer. Default: None.
@@ -1889,6 +2040,10 @@ class BilinearTensorProduct(layers.Layer):
            of this layer. If it is set to False, no bias will be added to the output units.
            If it is set to None, the bias is initialized zero. Default: None.
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
        Variable: A 2-D Tensor of shape [batch_size, size].
 
@@ -1940,6 +2095,22 @@ class BilinearTensorProduct(layers.Layer):
             dtype=self._dtype,
             is_bias=True)
 
+    @property
+    def weight(self):
+        return self._w
+
+    @weight.setter
+    def weight(self, value):
+        self._w = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, x, y):
         self._inputs = {"X": x, "Y": y, "Weight": self._w}
         if self._bias_param:
@@ -2013,7 +2184,7 @@ class Conv2DTranspose(layers.Layer):
            H_{out} &\in [ H^\prime_{out}, H^\prime_{out} + strides[0] ) \\\\
            W_{out} &\in [ W^\prime_{out}, W^\prime_{out} + strides[1] )
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         num_filters(int): The number of the filter. It is as same as the output
             image channel.
@@ -2055,6 +2226,10 @@ class Conv2DTranspose(layers.Layer):
         act (str): Activation type, if it is set to None, activation is not appended.
             Default: None.
 
+    Attributes:
+        weight (Parameter): the learnable weights of filters of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
         Variable: The tensor variable storing the convolution transpose result.
 
@@ -2163,6 +2338,22 @@ class Conv2DTranspose(layers.Layer):
             dtype=self._dtype,
             is_bias=True)
 
+    @property
+    def weight(self):
+        return self._img_filter
+
+    @weight.setter
+    def weight(self, value):
+        self._img_filter = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, input):
         pre_bias = self._helper.create_variable_for_type_inference(
             dtype=input.dtype)
@@ -2202,7 +2393,7 @@ class SequenceConv(layers.Layer):
     other convolutional configurations for the filters and stride as given
     in the input parameters to the function.
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         num_filters (int): number of filters.
         filter_size (int): the filter size (H and W). Default: 3.
@@ -2220,6 +2411,10 @@ class SequenceConv(layers.Layer):
         act (str): Activation type, if it is set to None, activation is not appended.
             Default: None.
 
+    Attributes:
+        weight (Parameter): the learnable weights of filters of this layer.
+        bias (Parameter|None): the learnable bias of this layer.
+
     Returns:
         Variable: output of sequence_conv
     """
@@ -2305,7 +2500,7 @@ class RowConv(layers.Layer):
 
     More details about row_conv please refer to the design document https://github.com/PaddlePaddle/Paddle/issues/2228#issuecomment-303903645 .
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         future_context_size (int): Future context size. Please note, the shape
             of convolution kernel is [future_context_size + 1, D].
@@ -2313,6 +2508,9 @@ class RowConv(layers.Layer):
             name, initializer etc. Default: None.
         act (str): Non-linear activation to be applied to output variable. Default: None.
 
+    Attributes:
+        weight (Parameter): the learnable weights of this layer.
+
     Returns:
         the output(Out) is a LodTensor, which supports variable time-length input sequences.
         The underlying tensor in this LodTensor is a matrix with shape T x N, i.e., the same shape as X.
@@ -2368,7 +2566,7 @@ class GroupNorm(layers.Layer):
 
         Refer to `Group Normalization <https://arxiv.org/abs/1803.08494>`_ .
 
-        Args:
+        Parameters:
             name_scope(str): The name of this class.
             groups(int): The number of groups that divided from channels.
             epsilon(float): The small value added to the variance to prevent
@@ -2496,7 +2694,7 @@ class SpectralNorm(layers.Layer):
 
     Refer to `Spectral Normalization <https://arxiv.org/abs/1802.05957>`_ .
 
-    Args:
+    Parameters:
         name_scope(str): The name of this class.
         dim(int): The index of dimension which should be permuted to the first before reshaping Input(Weight) to matrix, it should be set as 0 if Input(Weight) is the weight of fc layer, and should be set as 1 if Input(Weight) is the weight of conv layer. Default: 0.
         power_iters(int): The number of power iterations to calculate spectral norm. Default: 1.
@@ -2573,7 +2771,7 @@ class TreeConv(layers.Layer):
         The paper of Tree-Based Convolution Operator is here: https://arxiv.org/abs/1409.5718v1
 
 
-        Args:
+        Parameters:
             name_scope(str): The name of this class.
             output_size(int): output feature width
             num_filters(int): number of filters, Default: 1.
@@ -2583,6 +2781,10 @@ class TreeConv(layers.Layer):
             bias_attr(ParamAttr): the parameter attribute for the bias of this layer, Default: None.
             name(str): a name of this layer(optional). If set None, the layer will be named automatically, Default: None.
 
+        Attributes:
+            weight (Parameter): the learnable weights of filters of this layer.
+            bias (Parameter|None): the learnable bias of this layer.
+
         Returns:
             out(Variable): (Tensor) The feature vector of subtrees. The shape of the output tensor is [max_tree_node_size, output_size, num_filters]. The output tensor could be a new feature vector for next tree convolution layers
 
@@ -2639,6 +2841,22 @@ class TreeConv(layers.Layer):
             dtype=self._dtype,
             is_bias=False)
 
+    @property
+    def weight(self):
+        return self.W
+
+    @weight.setter
+    def weight(self, value):
+        self.W = value
+
+    @property
+    def bias(self):
+        return self._bias_param
+
+    @bias.setter
+    def bias(self, value):
+        self._bias_param = value
+
     def forward(self, nodes_vector, edge_set):
 
         if self._name:
diff --git a/python/paddle/fluid/framework.py b/python/paddle/fluid/framework.py
index 3b171ed5c68cbd73ed0d1fe6a6549f892c9bd995..e82103772213ccba70600749fab8b5eddc7eb2dd 100644
--- a/python/paddle/fluid/framework.py
+++ b/python/paddle/fluid/framework.py
@@ -638,6 +638,45 @@ class Variable(object):
         new_ivar = self._ivar._copy_to(core.CPUPlace(), True)
         return np.array(new_ivar.value().get_tensor())
 
+    @dygraph_only
+    def set_value(self, value):
+        """
+        Set a new value for this Variable.
+
+        Args:
+            value (Variable|np.ndarray): the new value.
+
+        Returns:
+            None.
+
+        Examples:
+            .. code-block:: python
+
+                import paddle.fluid as fluid
+                from paddle.fluid.dygraph.base import to_variable
+                from paddle.fluid.dygraph import FC
+                import numpy as np
+
+                data = np.ones([3, 32, 32], dtype='float32')
+                with fluid.dygraph.guard():
+                    fc = fluid.dygraph.FC("fc", 4)
+                    t = to_variable(data)
+                    fc(t)  # call with default weight
+                    custom_weight = np.random.randn(1024, 4).astype("float32")
+                    fc.weight.set_value(custom_weight)  # change existing weight
+                    out = fc(t)  # call with different weight
+
+        """
+        assert isinstance(value, (Variable, np.ndarray))
+        if list(value.shape) != list(self.shape):
+            raise ValueError(
+                "The shape of the new value must be the same as that of the original Variable."
+            )
+        self_tensor = self._ivar.value().get_tensor()
+        if isinstance(value, Variable):
+            value = value._ivar.value().get_tensor().__array__()
+        self_tensor.set(value, _current_expected_place())
+
     @dygraph_only
     def backward(self, backward_strategy=None):
         """
@@ -1042,7 +1081,7 @@ class Variable(object):
             if self.shape[axis] < 0:
                 return self._cloneVar(True)
             index = int(item)
-            if (index > 0 and index >= self.shape[axis])\
+            if (index > 0 and index >= self.shape[axis]) \
                     or (index < 0 and (index + self.shape[axis]) < 0):
                 raise IndexError("invalid index")
             return self._sliceVar([axis], [index], [index + 1])
@@ -2662,10 +2701,10 @@ class IrOpNode(IrNode):
         if isinstance(val, Block):
             desc.set_block_attr(name, val.desc)
         elif isinstance(val, list) and val and \
-            all(isinstance(v, Block) for v in val):
+                all(isinstance(v, Block) for v in val):
             desc.set_blocks_attr(name, [v.desc for v in val])
         elif isinstance(val, core.BlockDesc) or \
-            isinstance(val, core.ProgramDesc):
+                isinstance(val, core.ProgramDesc):
             desc.set_serialized_attr(name, val.serialize_to_string())
         else:
             desc._set_attr(name, val)
@@ -2888,8 +2927,8 @@ class IrGraph(object):
             op_node(IrOpNode): the operator node that is needed to update input's link.
         """
         assert old_input_node.node in self.graph.nodes() and new_input_node.node in \
-        self.graph.nodes() and op_node.node in self.graph.nodes(), \
-        'The three arguments(old_input_node&new_input_node&op_node) must be in the graph nodes.'
+               self.graph.nodes() and op_node.node in self.graph.nodes(), \
+            'The three arguments(old_input_node&new_input_node&op_node) must be in the graph nodes.'
         old_input_node.remove_output(op_node)
         op_node.remove_input(old_input_node)
         new_input_node.append_output(op_node)
@@ -3024,7 +3063,7 @@ class IrGraph(object):
         def _convert_to_pdf(dot_file_path):
             pdf_save_path = os.path.splitext(dot_file_path)[0] + '.pdf'
             exited_code = subprocess.call('dot -Tpdf ' + dot_file_path \
-                            + ' -o ' + pdf_save_path, shell=True)
+                                          + ' -o ' + pdf_save_path, shell=True)
             if exited_code != 0:
                 print('The dot command is needed for creating pdf files.')
                 print('The {} is saved as the dot filetype.'.format(
diff --git a/python/paddle/fluid/tests/unittests/test_layers.py b/python/paddle/fluid/tests/unittests/test_layers.py
index 838314115a7c804b61746cf947ceb31fcc83acba..920a212c6f7c1fb237ea83dd11f89ad717c61208 100644
--- a/python/paddle/fluid/tests/unittests/test_layers.py
+++ b/python/paddle/fluid/tests/unittests/test_layers.py
@@ -82,6 +82,34 @@ class LayerTest(unittest.TestCase):
 
 
 class TestLayer(LayerTest):
+    def test_custom_layer_with_kwargs(self):
+        class CustomLayer(fluid.Layer):
+            def __init__(self, name_scope, fc1_size=4):
+                super(CustomLayer, self).__init__(name_scope)
+                self.fc1 = nn.FC('fc1',
+                                 size=fc1_size,
+                                 bias_attr=False,
+                                 num_flatten_dims=1)
+                self.fc2 = nn.FC('fc2',
+                                 size=1,
+                                 bias_attr=False,
+                                 num_flatten_dims=1)
+
+            def forward(self, x, do_fc2=False):
+                ret = self.fc1(x)
+                if do_fc2:
+                    ret = self.fc2(ret)
+                return ret
+
+        with self.dynamic_graph():
+            inp = np.ones([3, 3], dtype='float32')
+            x = base.to_variable(inp)
+            custom = CustomLayer('custom', fc1_size=2)
+            ret = custom(x, do_fc2=False)
+            self.assertTrue(np.array_equal(ret.numpy().shape, [3, 2]))
+            ret = custom(x, do_fc2=True)
+            self.assertTrue(np.array_equal(ret.numpy().shape, [3, 1]))
+
     def test_fc(self):
         inp = np.ones([3, 32, 32], dtype='float32')
         with self.static_graph():
@@ -117,6 +145,62 @@ class TestLayer(LayerTest):
         self.assertTrue(np.array_equal(static_ret, static_ret2))
         self.assertTrue(np.array_equal(static_ret, dy_ret_value))
 
+        with self.dynamic_graph():
+            custom_weight = np.random.randn(1024, 4).astype("float32")
+            weight_attr1 = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            fc1 = fluid.dygraph.FC("fc1",
+                                   4,
+                                   num_flatten_dims=1,
+                                   param_attr=weight_attr1)
+            out1 = fc1(base.to_variable(inp))
+            loss1 = fluid.layers.reduce_mean(out1)
+
+            fc1_weight_init = fc1.weight.detach()
+            fc1_bias_init = fc1.bias.detach()
+
+            loss1.backward()
+            optimizer1 = fluid.optimizer.SGD(learning_rate=0.1)
+            optimizer1.minimize(loss1)
+
+            fc1_weight_updated = fc1.weight.detach()
+
+        with self.dynamic_graph():
+            weight_attr2 = fluid.ParamAttr(
+                initializer=fluid.initializer.Uniform())
+            fc2 = fluid.dygraph.FC("fc2",
+                                   4,
+                                   num_flatten_dims=1,
+                                   param_attr=weight_attr2)
+            out2 = fc2(base.to_variable(inp))
+
+            self.assertFalse(
+                np.array_equal(fc1_weight_init.numpy(), fc2.weight.numpy()))
+            self.assertFalse(np.array_equal(out1.numpy(), out2.numpy()))
+
+            mismatched_weight = np.random.randn(4, 4).astype("float32")
+            with self.assertRaises(ValueError):
+                fc2.weight.set_value(mismatched_weight)
+            fc2.weight.set_value(fc1_weight_init)
+            fc2.bias.set_value(fc1_bias_init)
+
+            out2 = fc2(base.to_variable(inp))
+            loss2 = fluid.layers.reduce_mean(out2)
+            loss2.backward()
+            optimizer2 = fluid.optimizer.SGD(learning_rate=0.1)
+            optimizer2.minimize(loss2)
+
+            self.assertTrue(
+                np.array_equal(fc2.weight.numpy(), fc1_weight_updated.numpy()))
+            self.assertTrue(np.array_equal(out1.numpy(), out2.numpy()))
+
+            fc2.weight = fc1.weight
+            fc2.bias = fc1.bias
+            self.assertTrue(
+                np.array_equal(fc2.weight.numpy(), fc1.weight.numpy()))
+            self.assertTrue(np.array_equal(fc2.bias.numpy(), fc1.bias.numpy()))
+
     def test_layer_norm(self):
         inp = np.ones([3, 32, 32], dtype='float32')
         with self.static_graph():
@@ -238,6 +322,41 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_ret, dy_ret_value))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
+        with self.dynamic_graph():
+            images = np.ones([2, 3, 5, 5], dtype='float32')
+            custom_weight = np.random.randn(3, 3, 2, 2).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            conv2d1 = nn.Conv2D('conv2d1', num_filters=3, filter_size=[2, 2])
+            conv2d2 = nn.Conv2D(
+                'conv2d2',
+                num_filters=3,
+                filter_size=[2, 2],
+                param_attr=weight_attr)
+            dy_ret1 = conv2d1(base.to_variable(images))
+            dy_ret2 = conv2d2(base.to_variable(images))
+            self.assertFalse(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv2d1_weight_np = conv2d1.weight.numpy()
+            conv2d1_bias = conv2d1.bias
+            self.assertFalse(
+                np.array_equal(conv2d1_weight_np, conv2d2.weight.numpy()))
+            conv2d2.weight.set_value(conv2d1_weight_np)
+            self.assertTrue(
+                np.array_equal(conv2d1_weight_np, conv2d2.weight.numpy()))
+            conv2d2.bias.set_value(conv2d1_bias)
+            dy_ret1 = conv2d1(base.to_variable(images))
+            dy_ret2 = conv2d2(base.to_variable(images))
+            self.assertTrue(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv2d2.weight = conv2d1.weight
+            conv2d2.bias = conv2d1.bias
+            self.assertTrue(
+                np.array_equal(conv2d1.weight.numpy(), conv2d2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(conv2d1.bias.numpy(), conv2d2.bias.numpy()))
+
     def test_gru_unit(self):
         lod = [[2, 4, 3]]
         D = 5
@@ -282,6 +401,37 @@ class TestLayer(LayerTest):
             self.assertTrue(np.allclose(static_ret[i], static_ret2[i]))
             self.assertTrue(np.allclose(static_ret[i], dy_ret_value[i]))
 
+        with self.dynamic_graph():
+            custom_weight = np.random.randn(D, D * 3).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            gru1 = nn.GRUUnit('gru1', size=D * 3)
+            gru2 = nn.GRUUnit('gru2', size=D * 3, param_attr=weight_attr)
+            dy_ret1 = gru1(
+                base.to_variable(input), base.to_variable(hidden_input))
+            dy_ret2 = gru2(
+                base.to_variable(input), base.to_variable(hidden_input))
+            self.assertFalse(
+                np.array_equal(gru1.weight.numpy(), gru2.weight.numpy()))
+            for o1, o2 in zip(dy_ret1, dy_ret2):
+                self.assertFalse(np.array_equal(o1.numpy(), o2.numpy()))
+            gru2.weight.set_value(gru1.weight.numpy())
+            gru2.bias.set_value(gru1.bias)
+            dy_ret1 = gru1(
+                base.to_variable(input), base.to_variable(hidden_input))
+            dy_ret2 = gru2(
+                base.to_variable(input), base.to_variable(hidden_input))
+            for o1, o2 in zip(dy_ret1, dy_ret2):
+                self.assertTrue(np.array_equal(o1.numpy(), o2.numpy()))
+
+            gru2.weight = gru1.weight
+            gru2.bias = gru1.bias
+            self.assertTrue(
+                np.array_equal(gru1.weight.numpy(), gru2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(gru1.bias.numpy(), gru2.bias.numpy()))
+
     def test_elementwise_math(self):
         n = np.ones([3, 3], dtype='float32')
         n2 = np.ones([3, 3], dtype='float32') * 1.1
@@ -417,6 +567,42 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
         self.assertTrue(np.allclose(dy_rlt_value, static_rlt2))
 
+        with self.dynamic_graph():
+            images = np.ones([2, 3, 5, 5], dtype='float32')
+            custom_weight = np.random.randn(3, 3, 2, 2).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            conv2d1 = nn.Conv2DTranspose(
+                'conv2d1', num_filters=3, filter_size=[2, 2])
+            conv2d2 = nn.Conv2DTranspose(
+                'conv2d2',
+                num_filters=3,
+                filter_size=[2, 2],
+                param_attr=weight_attr)
+            dy_ret1 = conv2d1(base.to_variable(images))
+            dy_ret2 = conv2d2(base.to_variable(images))
+            self.assertFalse(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv2d1_weight_np = conv2d1.weight.numpy()
+            conv2d1_bias = conv2d1.bias
+            self.assertFalse(
+                np.array_equal(conv2d1_weight_np, conv2d2.weight.numpy()))
+            conv2d2.weight.set_value(conv2d1_weight_np)
+            self.assertTrue(
+                np.array_equal(conv2d1_weight_np, conv2d2.weight.numpy()))
+            conv2d2.bias.set_value(conv2d1_bias)
+            dy_ret1 = conv2d1(base.to_variable(images))
+            dy_ret2 = conv2d2(base.to_variable(images))
+            self.assertTrue(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv2d2.weight = conv2d1.weight
+            conv2d2.bias = conv2d1.bias
+            self.assertTrue(
+                np.array_equal(conv2d1.weight.numpy(), conv2d2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(conv2d1.bias.numpy(), conv2d2.bias.numpy()))
+
     def test_bilinear_tensor_product(self):
         inp_np_x = np.array([[1, 2, 3]]).astype('float32')
         inp_np_y = np.array([[4, 5, 6]]).astype('float32')
@@ -498,9 +684,36 @@ class TestLayer(LayerTest):
         self.assertTrue(np.array_equal(static_rlt2, static_rlt))
         self.assertTrue(np.array_equal(dy_rlt_value, static_rlt))
 
+        with self.dynamic_graph():
+            custom_weight = np.random.randn(6, 3, 3).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            btp1 = nn.BilinearTensorProduct('btp1', 6, act='sigmoid')
+            btp2 = nn.BilinearTensorProduct(
+                'btp2', 6, act='sigmoid', param_attr=weight_attr)
+            dy_rlt1 = btp1(
+                base.to_variable(inp_np_x), base.to_variable(inp_np_y))
+            dy_rlt2 = btp2(
+                base.to_variable(inp_np_x), base.to_variable(inp_np_y))
+            self.assertFalse(np.array_equal(dy_rlt1.numpy(), dy_rlt2.numpy()))
+            btp2.weight.set_value(btp1.weight.numpy())
+            btp2.bias.set_value(btp1.bias)
+            dy_rlt1 = btp1(
+                base.to_variable(inp_np_x), base.to_variable(inp_np_y))
+            dy_rlt2 = btp2(
+                base.to_variable(inp_np_x), base.to_variable(inp_np_y))
+            self.assertTrue(np.array_equal(dy_rlt1.numpy(), dy_rlt2.numpy()))
+
+            btp2.weight = btp1.weight
+            btp2.bias = btp1.bias
+            self.assertTrue(
+                np.array_equal(btp1.weight.numpy(), btp2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(btp1.bias.numpy(), btp2.bias.numpy()))
+
     def test_prelu(self):
         inp_np = np.ones([5, 200, 100, 100]).astype('float32')
-
         with self.static_graph():
             data_t = layers.data(
                 name="input",
@@ -540,6 +753,32 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
         self.assertTrue(np.allclose(dy_rlt_value, static_rlt))
 
+        with self.dynamic_graph():
+            inp_np = np.random.randn(5, 200, 100, 100).astype("float32")
+            inp = base.to_variable(inp_np)
+            mode = 'channel'
+            prelu1 = nn.PRelu(
+                'prelu1',
+                mode=mode,
+                param_attr=ParamAttr(initializer=Constant(2.0)))
+            prelu2 = nn.PRelu(
+                'prelu2',
+                mode=mode,
+                param_attr=ParamAttr(initializer=Constant(1.0)))
+            dy_rlt1 = prelu1(inp)
+            dy_rlt2 = prelu2(inp)
+            self.assertFalse(
+                np.array_equal(prelu1.weight.numpy(), prelu2.weight.numpy()))
+            self.assertFalse(np.array_equal(dy_rlt1.numpy(), dy_rlt2.numpy()))
+            prelu2.weight.set_value(prelu1.weight.numpy())
+            dy_rlt1 = prelu1(inp)
+            dy_rlt2 = prelu2(inp)
+            self.assertTrue(np.array_equal(dy_rlt1.numpy(), dy_rlt2.numpy()))
+
+            prelu2.weight = prelu1.weight
+            self.assertTrue(
+                np.array_equal(prelu1.weight.numpy(), prelu2.weight.numpy()))
+
     def test_embeding(self):
         inp_word = np.array([[[1]]]).astype('int64')
         dict_size = 20
@@ -574,6 +813,31 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
         self.assertTrue(np.allclose(dy_rlt_value, static_rlt))
 
+        with self.dynamic_graph():
+            custom_weight = np.random.randn(dict_size, 32).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            emb1 = nn.Embedding(
+                name_scope='embedding', size=[dict_size, 32], is_sparse=False)
+            emb2 = nn.Embedding(
+                name_scope='embedding',
+                size=[dict_size, 32],
+                param_attr=weight_attr,
+                is_sparse=False)
+            rep1 = emb1(base.to_variable(inp_word))
+            rep2 = emb2(base.to_variable(inp_word))
+            self.assertFalse(np.array_equal(emb1.weight.numpy(), custom_weight))
+            self.assertTrue(np.array_equal(emb2.weight.numpy(), custom_weight))
+            self.assertFalse(np.array_equal(rep1.numpy(), rep2.numpy()))
+            emb2.weight.set_value(emb1.weight.numpy())
+            rep2 = emb2(base.to_variable(inp_word))
+            self.assertTrue(np.array_equal(rep1.numpy(), rep2.numpy()))
+
+            emb2.weight = emb1.weight
+            self.assertTrue(
+                np.array_equal(emb1.weight.numpy(), emb2.weight.numpy()))
+
     def test_nce(self):
         window_size = 5
         dict_size = 20
@@ -695,6 +959,69 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
         self.assertTrue(np.allclose(dy_rlt_value, static_rlt))
 
+        with self.dynamic_graph(force_to_use_cpu=True):
+            custom_weight = np.random.randn(dict_size, 128).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            words = []
+            for i in range(window_size):
+                words.append(base.to_variable(inp_word[i]))
+            sample_weights = layers.fill_constant(
+                shape=[5, 1], dtype='float32', value=1)
+            emb = nn.Embedding(
+                'embedding',
+                size=[dict_size, 32],
+                param_attr='emb.w',
+                is_sparse=False)
+
+            embs3 = []
+            for i in range(window_size):
+                if i == label_word:
+                    continue
+
+                emb_rlt = emb(words[i])
+                embs3.append(emb_rlt)
+
+            embs3 = layers.concat(input=embs3, axis=1)
+            nce1 = nn.NCE('nce1',
+                          num_total_classes=dict_size,
+                          num_neg_samples=2,
+                          sampler="custom_dist",
+                          custom_dist=nid_freq_arr.tolist(),
+                          seed=seed,
+                          param_attr='nce1.w',
+                          bias_attr='nce1.b',
+                          sample_weight=sample_weights)
+
+            nce2 = nn.NCE('nce2',
+                          param_attr=weight_attr,
+                          num_total_classes=dict_size,
+                          num_neg_samples=2,
+                          sampler="custom_dist",
+                          custom_dist=nid_freq_arr.tolist(),
+                          seed=seed,
+                          bias_attr='nce2.b',
+                          sample_weight=sample_weights)
+
+            nce1_loss = nce1(embs3, words[label_word])
+            nce2_loss = nce2(embs3, words[label_word])
+            self.assertFalse(
+                np.array_equal(nce1_loss.numpy(), nce2_loss.numpy()))
+            nce2.weight.set_value(nce1.weight.numpy())
+            nce2.bias.set_value(nce1.bias)
+            nce1_loss = nce1(embs3, words[label_word])
+            nce2_loss = nce2(embs3, words[label_word])
+            self.assertTrue(
+                np.array_equal(nce1_loss.numpy(), nce2_loss.numpy()))
+
+            nce2.weight = nce1.weight
+            nce2.bias = nce1.bias
+            self.assertTrue(
+                np.array_equal(nce1.weight.numpy(), nce2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(nce1.bias.numpy(), nce2.bias.numpy()))
+
     def test_conv3d(self):
         with self.static_graph():
             images = layers.data(
@@ -724,6 +1051,38 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_ret, dy_rlt_value))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
+        with self.dynamic_graph():
+            images = np.ones([2, 3, 6, 6, 6], dtype='float32')
+            custom_weight = np.random.randn(3, 3, 2, 2, 2).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            conv3d1 = nn.Conv3D('conv3d1', num_filters=3, filter_size=2)
+            conv3d2 = nn.Conv3D(
+                'conv3d2', num_filters=3, filter_size=2, param_attr=weight_attr)
+            dy_ret1 = conv3d1(base.to_variable(images))
+            dy_ret2 = conv3d2(base.to_variable(images))
+            self.assertFalse(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv3d1_weight_np = conv3d1.weight.numpy()
+            conv3d1_bias = conv3d1.bias
+            self.assertFalse(
+                np.array_equal(conv3d1_weight_np, conv3d2.weight.numpy()))
+            conv3d2.weight.set_value(conv3d1_weight_np)
+            self.assertTrue(
+                np.array_equal(conv3d1_weight_np, conv3d2.weight.numpy()))
+            conv3d1.bias.set_value(conv3d1_bias)
+            dy_ret1 = conv3d1(base.to_variable(images))
+            dy_ret2 = conv3d2(base.to_variable(images))
+            self.assertTrue(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv3d2.weight = conv3d1.weight
+            conv3d2.bias = conv3d1.bias
+            self.assertTrue(
+                np.array_equal(conv3d1.weight.numpy(), conv3d2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(conv3d1.bias.numpy(), conv3d2.bias.numpy()))
+
     def test_row_conv(self):
         input = np.arange(15).reshape([3, 5]).astype('float32')
         if core.is_compiled_with_cuda():
@@ -943,6 +1302,45 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_ret, static_ret2))
         self.assertTrue(np.allclose(static_ret, dy_rlt_value))
 
+        with self.dynamic_graph():
+            custom_weight = np.random.randn(5, 3, 6, 1).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            treeConv1 = nn.TreeConv(
+                'SpectralNorm1',
+                output_size=6,
+                num_filters=1,
+                max_depth=2,
+                bias_attr='tc1_b')
+            treeConv2 = nn.TreeConv(
+                'SpectralNorm2',
+                output_size=6,
+                num_filters=1,
+                max_depth=2,
+                param_attr=weight_attr,
+                bias_attr='tc2_b')
+            dy_ret1 = treeConv1(
+                base.to_variable(vectors), base.to_variable(adj))
+            dy_ret2 = treeConv2(
+                base.to_variable(vectors), base.to_variable(adj))
+            self.assertFalse(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+            treeConv2.weight.set_value(treeConv1.weight.numpy())
+            treeConv2.bias.set_value(treeConv1.bias)
+            dy_ret1 = treeConv1(
+                base.to_variable(vectors), base.to_variable(adj))
+            dy_ret2 = treeConv2(
+                base.to_variable(vectors), base.to_variable(adj))
+            self.assertTrue(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            treeConv2.weight = treeConv1.weight
+            treeConv2.bias = treeConv1.bias
+            self.assertTrue(
+                np.array_equal(treeConv1.weight.numpy(),
+                               treeConv2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(treeConv1.bias.numpy(), treeConv2.bias.numpy()))
+
     def test_conv3d_transpose(self):
         input_array = np.arange(0, 48).reshape(
             [2, 3, 2, 2, 2]).astype('float32')
@@ -974,6 +1372,48 @@ class TestLayer(LayerTest):
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
         self.assertTrue(np.allclose(dy_rlt_value, static_rlt))
 
+        with self.dynamic_graph():
+            images = np.ones([2, 3, 6, 6, 6], dtype='float32')
+            custom_weight = np.random.randn(3, 3, 2, 2, 2).astype("float32")
+            weight_attr = fluid.ParamAttr(
+                initializer=fluid.initializer.NumpyArrayInitializer(
+                    custom_weight))
+            conv3d1 = nn.Conv3DTranspose(
+                'conv3d1',
+                num_filters=3,
+                filter_size=2,
+                bias_attr='conv3d1_b',
+                use_cudnn=False)
+            conv3d2 = nn.Conv3DTranspose(
+                'conv3d2',
+                num_filters=3,
+                filter_size=2,
+                param_attr=weight_attr,
+                bias_attr='conv3d2_b',
+                use_cudnn=False)
+            dy_ret1 = conv3d1(base.to_variable(images))
+            dy_ret2 = conv3d2(base.to_variable(images))
+            self.assertFalse(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv3d1_weight_np = conv3d1.weight.numpy()
+            conv3d1_bias = conv3d1.bias
+            self.assertFalse(
+                np.array_equal(conv3d1_weight_np, conv3d2.weight.numpy()))
+            conv3d2.weight.set_value(conv3d1_weight_np)
+            self.assertTrue(
+                np.array_equal(conv3d1_weight_np, conv3d2.weight.numpy()))
+            conv3d1.bias.set_value(conv3d1_bias)
+            dy_ret1 = conv3d1(base.to_variable(images))
+            dy_ret2 = conv3d2(base.to_variable(images))
+            self.assertTrue(np.array_equal(dy_ret1.numpy(), dy_ret2.numpy()))
+
+            conv3d2.weight = conv3d1.weight
+            conv3d2.bias = conv3d1.bias
+            self.assertTrue(
+                np.array_equal(conv3d1.weight.numpy(), conv3d2.weight.numpy()))
+            self.assertTrue(
+                np.array_equal(conv3d1.bias.numpy(), conv3d2.bias.numpy()))
+
     def test_eye_op(self):
         np_eye = np.eye(3, 2)
         array_rlt1 = [np_eye for _ in range(3)]