Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into sub_sequence_op

.gitignore

@@ -21,7 +21,7 @@ third_party/
 cmake-build-*
 
 # generated while compiling
-python/paddle/v2/framework/core.so
+python/paddle/v2/fluid/core.so
 paddle/pybind/pybind.h
 CMakeFiles
 cmake_install.cmake

paddle/operators/beam_search_decode_op.cc

@@ -27,6 +27,7 @@ class BeamSearchDecodeOp : public framework::OperatorBase {
   void Run(const framework::Scope& scope,
            const platform::DeviceContext& dev_ctx) const override {
     framework::ExecutionContext ctx(*this, scope, dev_ctx);
+
     const LoDTensorArray* ids = ctx.Input<LoDTensorArray>("Ids");
     const LoDTensorArray* scores = ctx.Input<LoDTensorArray>("Scores");
     const size_t step_num = ids->size();

paddle/operators/l1_norm_op.h

@@ -29,7 +29,7 @@ class L1NormKernel : public framework::OpKernel<T> {
     Out->mutable_data<T>(context.GetPlace());
 
     auto x = framework::EigenVector<T>::Flatten(*X);
-    auto out = framework::EigenVector<T>::Flatten(*Out);
+    auto out = framework::EigenScalar<T>::From(*Out);
     auto place = context.GetEigenDevice<Place>();
 
     out.device(place) = x.abs().sum();

paddle/operators/lod_reset_op.cc

@@ -0,0 +1,120 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#include "paddle/operators/lod_reset_op.h"
+
+namespace paddle {
+namespace operators {
+
+class LoDResetOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+  void InferShape(framework::InferShapeContext *ctx) const override {
+    // input check
+    PADDLE_ENFORCE(ctx->HasInput("X"),
+                   "Input(X) of LoDResetOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasOutput("Out"),
+                   "Output(Out) of LoDResetOp should not be null.");
+    // If target LoD is not set form Input(), then it must be set from Attr().
+    if (!ctx->HasInput("TargetLoD")) {
+      auto level0 = ctx->Attrs().Get<std::vector<int>>("target_lod");
+      PADDLE_ENFORCE(level0.size() > 1,
+                     "Target LoD is not found, should be set to be a valid one "
+                     "through Input() or Attr().");
+    }
+    ctx->SetOutputDim("Out", ctx->GetInputDim("X"));
+  }
+
+ protected:
+  framework::OpKernelType GetKernelType(
+      const framework::ExecutionContext &ctx) const override {
+    return framework::OpKernelType(
+        framework::ToDataType(ctx.Input<framework::LoDTensor>("X")->type()),
+        ctx.device_context());
+  }
+};
+
+class LoDResetOpMaker : public framework::OpProtoAndCheckerMaker {
+ public:
+  LoDResetOpMaker(framework::OpProto *proto,
+                  framework::OpAttrChecker *op_checker)
+      : OpProtoAndCheckerMaker(proto, op_checker) {
+    AddInput("X", "(LoDTensor) The input tensor of lod_reset operator.");
+    AddInput("TargetLoD",
+             "(Tensor, optional) The target level 0 LoD from Input().")
+        .AsDispensable();
+    AddOutput("Out", "(LoDTensor) The output tensor of lod_reset operator.");
+    AddAttr<std::vector<int>>("target_lod",
+                              "The target level 0 LoD from Attr().")
+        .SetDefault(std::vector<int>{});
+    AddComment(R"DOC(LoDReset operator
+
+Reset LoD of Input(X) into a new one specified by Input(TargetLoD) or
+Attr(target_lod), or set LoD for Input(X) if it doesn't have one.
+Currently the lod_reset operator only supports the reset of level 0 LoD.
+At least one of Input(TargetLoD) and Attr(target_lod) must be set,
+and if both of them are set, Input(TargetLoD) will be chosen as the
+target LoD.
+
+An example:
+Given a float LoDTensor X with shape (6, 1), its transpose form represents
+
+    [1.0, 2.0, 3.0, 4.0, 5.0, 6.0],
+
+with LoD = [[0, 2, 5, 6]] and the three (transposed) sequences look like
+
+    [1.0, 2.0], [3.0, 4.0, 5.0], [6.0].
+
+If target LoD = [0, 4, 6], the lod_reset operator will reset the LoD and
+the sequences that the LoDTensor Output(Out) contains becomes:
+
+    [1.0, 2.0, 3.0, 4.0], [5.0, 6.0].
+
+)DOC");
+  }
+};
+
+class LoDResetGradOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+  void InferShape(framework::InferShapeContext *ctx) const override {
+    PADDLE_ENFORCE(ctx->HasInput("X"), "Input(X) shouldn't be null.");
+    PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Out")),
+                   "Input(Out@GRAD) shouldn't be null.");
+    ctx->SetOutputDim(framework::GradVarName("X"), ctx->GetInputDim("X"));
+  }
+
+ protected:
+  framework::OpKernelType GetKernelType(
+      const framework::ExecutionContext &ctx) const override {
+    return framework::OpKernelType(
+        framework::ToDataType(ctx.Input<framework::LoDTensor>("X")->type()),
+        ctx.device_context());
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP(lod_reset, ops::LoDResetOp, ops::LoDResetOpMaker, lod_reset_grad,
+            ops::LoDResetGradOp);
+REGISTER_OP_CPU_KERNEL(lod_reset,
+                       ops::LoDResetKernel<paddle::platform::CPUPlace, float>,
+                       ops::LoDResetKernel<paddle::platform::CPUPlace, double>);
+REGISTER_OP_CPU_KERNEL(
+    lod_reset_grad, ops::LoDResetGradKernel<paddle::platform::CPUPlace, float>,
+    ops::LoDResetGradKernel<paddle::platform::CPUPlace, double>);

paddle/operators/lod_reset_op.cu

@@ -0,0 +1,24 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#include "paddle/operators/lod_reset_op.h"
+
+namespace ops = paddle::operators;
+
+REGISTER_OP_GPU_KERNEL(lod_reset,
+                       ops::LoDResetKernel<paddle::platform::GPUPlace, float>,
+                       ops::LoDResetKernel<paddle::platform::GPUPlace, double>);
+REGISTER_OP_GPU_KERNEL(
+    lod_reset_grad, ops::LoDResetGradKernel<paddle::platform::GPUPlace, float>,
+    ops::LoDResetGradKernel<paddle::platform::GPUPlace, double>);

paddle/operators/lod_reset_op.h

@@ -0,0 +1,78 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#pragma once
+
+#include "paddle/framework/eigen.h"
+#include "paddle/framework/op_registry.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename Place, typename T>
+class LoDResetKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const {
+    auto* out = ctx.Output<framework::LoDTensor>("Out");
+    auto* in = ctx.Input<framework::LoDTensor>("X");
+    auto* lod_t = ctx.Input<framework::Tensor>("TargetLoD");
+
+    std::vector<int> level0;
+    if (lod_t) {
+      auto* lod = lod_t->data<int>();
+      if (platform::is_gpu_place(ctx.GetPlace())) {
+        framework::Tensor lod_cpu;
+        lod_cpu.CopyFrom(*lod_t, platform::CPUPlace(), ctx.device_context());
+        lod = lod_cpu.data<int>();
+      }
+      level0 = std::vector<int>(lod, lod + lod_t->numel());
+    } else {
+      level0 = ctx.Attr<std::vector<int>>("target_lod");
+    }
+
+    PADDLE_ENFORCE(level0.size() > 1UL,
+                   "The size of target LoD should be greater than 1.");
+    PADDLE_ENFORCE(level0[0] == 0,
+                   "Target LoD should be a vector starting from 0.");
+    PADDLE_ENFORCE(level0.back() == in->dims()[0],
+                   "Target LoD should be a vector end with the "
+                   "first dimension of Input(X).");
+    for (size_t i = 0; i < level0.size() - 1; ++i) {
+      PADDLE_ENFORCE(level0[i + 1] > level0[i],
+                     "Target LoD should be an ascending vector.");
+    }
+
+    out->ShareDataWith(*in);
+    // cast level0 to size_t
+    std::vector<size_t> ulevel0(level0.size(), 0);
+    std::transform(level0.begin(), level0.end(), ulevel0.begin(),
+                   [](int a) { return static_cast<size_t>(a); });
+    framework::LoD target_lod;
+    target_lod.push_back(ulevel0);
+    out->set_lod(target_lod);
+  }
+};
+
+template <typename Place, typename T>
+class LoDResetGradKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const {
+    auto* d_out = ctx.Input<framework::Tensor>(framework::GradVarName("Out"));
+    auto* d_x = ctx.Output<framework::Tensor>(framework::GradVarName("X"));
+
+    d_x->ShareDataWith(*d_out);
+  }
+};
+}  // namespace operators
+}  // namespace paddle

paddle/operators/squared_l2_norm_op.h

@@ -29,7 +29,7 @@ class SquaredL2NormKernel : public framework::OpKernel<T> {
     Out->mutable_data<T>(context.GetPlace());
 
     auto x = framework::EigenVector<T>::Flatten(*X);
-    auto out = framework::EigenVector<T>::Flatten(*Out);
+    auto out = framework::EigenScalar<T>::From(*Out);
     auto place = context.GetEigenDevice<Place>();
 
     out.device(place) = x.square().sum();

paddle/pybind/pybind.cc

@@ -42,6 +42,9 @@ limitations under the License. */
 #include "paddle/platform/gpu_info.h"
 #endif
 
+// disable auto conversion to list in Python
+PYBIND11_MAKE_OPAQUE(paddle::framework::LoDTensorArray);
+
 namespace paddle {
 namespace pybind {
 static size_t UniqueIntegerGenerator(const std::string &prefix) {

python/CMakeLists.txt

@@ -37,10 +37,10 @@ configure_file(${CMAKE_CURRENT_SOURCE_DIR}/setup.py.in
     ${CMAKE_CURRENT_BINARY_DIR}/setup.py)
 
 
-add_custom_command(OUTPUT ${PADDLE_SOURCE_DIR}/python/paddle/v2/framework/core.so
+add_custom_command(OUTPUT ${PADDLE_SOURCE_DIR}/python/paddle/v2/fluid/core.so
-        COMMAND cmake -E copy $<TARGET_FILE:paddle_pybind> ${PADDLE_SOURCE_DIR}/python/paddle/v2/framework/core.so
+        COMMAND cmake -E copy $<TARGET_FILE:paddle_pybind> ${PADDLE_SOURCE_DIR}/python/paddle/v2/fluid/core.so
         DEPENDS paddle_pybind)
-add_custom_target(copy_paddle_pybind ALL DEPENDS ${PADDLE_SOURCE_DIR}/python/paddle/v2/framework/core.so)
+add_custom_target(copy_paddle_pybind ALL DEPENDS ${PADDLE_SOURCE_DIR}/python/paddle/v2/fluid/core.so)
 
 
 add_custom_command(OUTPUT ${PADDLE_PYTHON_BUILD_DIR}/.timestamp
@@ -66,7 +66,7 @@ if (WITH_TESTING)
     add_subdirectory(paddle/v2/tests)
     add_subdirectory(paddle/v2/reader/tests)
     add_subdirectory(paddle/v2/plot/tests)
-    add_subdirectory(paddle/v2/framework/tests)
+    add_subdirectory(paddle/v2/fluid/tests)
   endif()
 endif()
 install(DIRECTORY ${PADDLE_PYTHON_PACKAGE_DIR}

python/paddle/trainer_config_helpers/layers.py

@@ -3592,10 +3592,9 @@ def lstm_step_layer(input,
     :type gate_act: BaseActivation
     :param state_act: State Activation Type. TanhActivation is the default.
     :type state_act: BaseActivation
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                     set to True or None, the bias is initialized to zero.
-                      parameter is set to True, the bias is initialized to zero.
+    :type bias_attr: ParameterAttribute | None | True
-    :type bias_attr: ParameterAttribute | None | bool | Any
     :param layer_attr: layer's extra attribute.
     :type layer_attr: ExtraLayerAttribute
     :return: LayerOutput object.
@@ -3650,9 +3649,10 @@ def gru_step_layer(input,
     :param name: The name of this layer. It is optional.
     :param gate_act: Activation type of this layer's two gates. Default is Sigmoid.
     :type gate_act: BaseActivation
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                      False or an object whose type is not ParameterAttribute, no bias
-                      parameter is set to True, the bias is initialized to zero.
+                      is defined. If this parameter is set to True,
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param param_attr: the parameter_attribute for transforming the output_mem
                        from previous step.
@@ -3712,9 +3712,10 @@ def gru_step_naive_layer(input,
     :type act: BaseActivation
     :param gate_act: Activation type of this layer's two gates. Default is Sigmoid.
     :type gate_act: BaseActivation
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                      False or an object whose type is not ParameterAttribute, no bias
-                      parameter is set to True, the bias is initialized to zero.
+                      is defined. If this parameter is set to True,
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param param_attr:
     :param layer_attr:
@@ -3844,9 +3845,10 @@ def recurrent_layer(input,
     :type input: LayerOutput
     :param act: Activation type. TanhActivation is the default.
     :type act: BaseActivation
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to 
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                      False or an object whose type is not ParameterAttribute,
-                      parameter is set to True, the bias is initialized to zero.
+                      no bias is defined. If the parameter is set to True,
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param param_attr: parameter attribute.
     :type param_attr: ParameterAttribute
@@ -4836,9 +4838,10 @@ def tensor_layer(a,
     :type act: BaseActivation
     :param param_attr: The Parameter Attribute.
     :type param_attr: ParameterAttribute
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                      False or an object whose type is not ParameterAttribute,
-                      parameter is set to True, the bias is initialized to zero.
+                      no bias is defined. If this parameter is set to True,
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param layer_attr: Extra Layer config.
     :type layer_attr: ExtraLayerAttribute | None
@@ -4900,9 +4903,10 @@ def selective_fc_layer(input,
     :type act: BaseActivation
     :param param_attr: The Parameter Attribute.
     :type param_attr: ParameterAttribute
-    :param bias_attr: The bias attribute. If the parameter is set to False or an object
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to
-                      whose type is not ParameterAttribute, no bias is defined. If the
+                      False or an object whose type is not ParameterAttribute,
-                      parameter is set to True, the bias is initialized to zero.
+                      no bias is defined. If this parameter is set to True,
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param layer_attr: Extra Layer config.
     :type layer_attr: ExtraLayerAttribute | None
@@ -5585,10 +5589,10 @@ def nce_layer(input,
                              to the num_classes. Each member of the list defines
                              the probability of a class given input x.
     :type neg_distribution: list | tuple | collections.Sequence | None
-    :param bias_attr: The attribute for bias. If this parameter is set False or
+    :param bias_attr: The parameter attribute for bias. If this parameter is set to
-                      any object whose type is not ParameterAttribute, no bias
+                      False or an object whose type is not ParameterAttribute,
-                      is added. If this parameter is set True, the bias is
+                      no bias is defined. If this parameter is set to True,
-                      initialized to zero.
+                      the bias is initialized to zero.
     :type bias_attr: ParameterAttribute | None | bool | Any
     :param layer_attr: Extra Layer Attribute.
     :type layer_attr: ExtraLayerAttribute
@@ -6498,9 +6502,9 @@ def gated_unit_layer(input,
     :param gate_param_attr: The parameter attribute of the gate. See ParameterAttribute
                             for details.
     :type gate_param_attr: ParameterAttribute
-    :param gate_bias_attr: The bias attribute of the gate. If the parameter is set to False or
+    :param gate_bias_attr: The bias attribute of the gate. If this parameter is set to False or
                            an object whose type is not ParameterAttribute, no bias is defined.
-                           If the parameter is set to True, the bias is initialized to zero.
+                           If this parameter is set to True, the bias is initialized to zero.
     :type gate_bias_attr: ParameterAttribute | bool | None | Any
     :param inproj_attr: Extra layer attributes of the projection. See ExtraLayerAttribute for
                         details.
@@ -6508,9 +6512,9 @@ def gated_unit_layer(input,
     :param inproj_param_attr: The parameter attribute of the projection. See ParameterAttribute
                               for details.
     :type inproj_param_attr: ParameterAttribute
-    :param inproj_bias_attr: The bias attribute of the projection. If the parameter is set to False
+    :param inproj_bias_attr: The bias attribute of the projection. If this parameter is set to False
                              or an object whose type is not ParameterAttribute, no bias is defined.
-                             If the parameter is set to True, the bias is initialized to zero.
+                             If this parameter is set to True, the bias is initialized to zero.
     :type inproj_bias_attr: ParameterAttribute | bool | None | Any
     :param layer_attr: Extra layer attribute of the product. See ExtraLayerAttribute for
                        details.

python/paddle/trainer_config_helpers/networks.py

@@ -681,34 +681,42 @@ def lstmemory_unit(input,
                                    state_act=TanhActivation())
 
 
-    :param input: input layer.
+    :param input: Input layer.
     :type input: LayerOutput
-    :param out_memory: output of previous time step
+    :param out_memory: The output of previous time step.
     :type out_memory: LayerOutput | None
-    :param name: lstmemory unit name.
+    :param name: The lstmemory unit name.
     :type name: basestring
-    :param size: lstmemory unit size.
+    :param size: The lstmemory unit size.
     :type size: int
-    :param param_attr: parameter attribute, None means default attribute.
+    :param param_attr: The parameter attribute for the weights in
+                     input to hidden projection.
+                     None means default attribute.
     :type param_attr: ParameterAttribute
-    :param act: last activiation type of lstm.
+    :param act: The last activiation type of lstm.
     :type act: BaseActivation
-    :param gate_act: gate activiation type of lstm.
+    :param gate_act: The gate activiation type of lstm.
     :type gate_act: BaseActivation
-    :param state_act: state activiation type of lstm.
+    :param state_act: The state activiation type of lstm.
     :type state_act: BaseActivation
-    :param input_proj_bias_attr: bias attribute for input to hidden projection.
+    :param input_proj_bias_attr: The parameter attribute for the bias in
-                False means no bias, None means default bias.
+                      input to hidden projection.
-    :type input_proj_bias_attr: ParameterAttribute|False|None
+                      False or None means no bias.
-    :param input_proj_layer_attr: extra layer attribute for input to hidden
+                      If this parameter is set to True,
-                projection of the LSTM unit, such as dropout, error clipping.
+                      the bias is initialized to zero.
+    :type input_proj_bias_attr: ParameterAttribute|bool|None
+    :param input_proj_layer_attr: The extra layer attribute for
+                     input to hidden projection of the LSTM unit,
+                     such as dropout, error clipping.
     :type input_proj_layer_attr: ExtraLayerAttribute
-    :param lstm_bias_attr: bias parameter attribute of lstm layer.
+    :param lstm_bias_attr: The parameter attribute for the bias in lstm layer.
-                False means no bias, None means default bias.
+                      False or None means no bias.
-    :type lstm_bias_attr: ParameterAttribute|False|None
+                      If this parameter is set to True,
-    :param lstm_layer_attr: extra attribute of lstm layer.
+                      the bias is initialized to zero.
+    :type lstm_bias_attr: ParameterAttribute|True|None
+    :param lstm_layer_attr: The extra attribute of lstm layer.
     :type lstm_layer_attr: ExtraLayerAttribute
-    :return: lstmemory unit name.
+    :return: The lstmemory unit name.
     :rtype: LayerOutput
     """
     if size is None:
@@ -786,34 +794,42 @@ def lstmemory_group(input,
                                     gate_act=SigmoidActivation(),
                                     state_act=TanhActivation())
 
-    :param input: input layer.
+    :param input: Input layer.
     :type input: LayerOutput
-    :param size: lstmemory group size.
+    :param size: The lstmemory group size.
     :type size: int
-    :param name: name of lstmemory group.
+    :param name: The name of lstmemory group.
     :type name: basestring
-    :param out_memory: output of previous time step.
+    :param out_memory: The output of previous time step.
     :type out_memory: LayerOutput | None
-    :param reverse: process the input in a reverse order or not.
+    :param reverse: Process the input in a reverse order or not.
     :type reverse: bool
-    :param param_attr: parameter attribute, None means default attribute.
+    :param param_attr: The parameter attribute for the weights in
+                     input to hidden projection.
+                     None means default attribute.
     :type param_attr: ParameterAttribute
-    :param act: last activiation type of lstm.
+    :param act: The last activiation type of lstm.
     :type act: BaseActivation
-    :param gate_act: gate activiation type of lstm.
+    :param gate_act: The gate activiation type of lstm.
     :type gate_act: BaseActivation
-    :param state_act: state activiation type of lstm.
+    :param state_act: The state activiation type of lstm.
     :type state_act: BaseActivation
-    :param lstm_bias_attr: bias parameter attribute of lstm layer.
+    :param input_proj_bias_attr: The parameter attribute for the bias in
-                           False means no bias, None means default bias.
+                      input to hidden projection.
-    :type lstm_bias_attr: ParameterAttribute|False|None
+                      False or None means no bias.
-    :param input_proj_bias_attr: bias attribute for input to hidden projection.
+                      If this parameter is set to True,
-                False means no bias, None means default bias.
+                      the bias is initialized to zero.
-    :type input_proj_bias_attr: ParameterAttribute|False|None
+    :type input_proj_bias_attr: ParameterAttribute|bool|None
-    :param input_proj_layer_attr: extra layer attribute for input to hidden
+    :param input_proj_layer_attr: The extra layer attribute for
-                projection of the LSTM unit, such as dropout, error clipping.
+                     input to hidden projection of the LSTM unit,
+                     such as dropout, error clipping.
     :type input_proj_layer_attr: ExtraLayerAttribute
-    :param lstm_layer_attr: lstm layer's extra attribute.
+    :param lstm_bias_attr: The parameter attribute for the bias in lstm layer.
+                      False or None means no bias.
+                      If this parameter is set to True,
+                      the bias is initialized to zero.
+    :type lstm_bias_attr: ParameterAttribute|True|None
+    :param lstm_layer_attr: The extra attribute of lstm layer.
     :type lstm_layer_attr: ExtraLayerAttribute
     :return: the lstmemory group.
     :rtype: LayerOutput

python/paddle/v2/fluid/backward.py

@@ -1,4 +1,4 @@
-from paddle.v2.framework import framework as framework
+from paddle.v2.fluid import framework as framework
 
 __all__ = ['append_backward_ops']
 

python/paddle/v2/fluid/default_scope_funcs.py

@@ -13,7 +13,7 @@ A `scoped_function` will take a `function` as input. That function will be
 invoked in a new local scope. 
 """
 
-import paddle.v2.framework.core
+import paddle.v2.fluid.core
 import threading
 
 __tl_scope__ = threading.local()
@@ -27,13 +27,13 @@ __all__ = [
 def get_cur_scope():
     """
     Get current scope.
-    :rtype: paddle.v2.framework.core.Scope
+    :rtype: paddle.v2.fluid.core.Scope
     """
     cur_scope_stack = getattr(__tl_scope__, 'cur_scope', None)
     if cur_scope_stack is None:
         __tl_scope__.cur_scope = list()
     if len(__tl_scope__.cur_scope) == 0:
-        __tl_scope__.cur_scope.append(paddle.v2.framework.core.Scope())
+        __tl_scope__.cur_scope.append(paddle.v2.fluid.core.Scope())
     return __tl_scope__.cur_scope[-1]
 
 

python/paddle/v2/fluid/evaluator.py

@@ -1,6 +1,6 @@
-import paddle.v2.framework.op as op
+import paddle.v2.fluid.op as op
 import numpy as np
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
 
 
 def avg_accumulate(accumulated_var, per_eval, num_batches, place):
@@ -22,7 +22,7 @@ class Evaluator(object):
         NOTE: default run on CPUPlace(), running on GPUPlace doesn't improve performance much.
 
         :param scope: the scope instance contains the input.
-        :type scope: paddle.v2.framework.core.scope
+        :type scope: paddle.v2.fluid.core.scope
         :param operator: operator name for caculating the evaluation for each mini-batch.
         :type operator: string
         :param input: output variable name of forward network.

python/paddle/v2/fluid/executor.py

@@ -1,5 +1,5 @@
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
-from paddle.v2.framework.framework import Block, Program, g_main_program
+from paddle.v2.fluid.framework import Block, Program, g_main_program
 
 g_scope = core.Scope()
 

python/paddle/v2/fluid/framework.py

@@ -1,5 +1,5 @@
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
-import paddle.v2.framework.proto.framework_pb2 as framework_pb2
+import paddle.v2.fluid.proto.framework_pb2 as framework_pb2
 import collections
 import numpy as np
 import copy

python/paddle/v2/fluid/initializer.py

@@ -1,4 +1,4 @@
-import paddle.v2.framework.framework as framework
+import paddle.v2.fluid.framework as framework
 import numpy as np
 
 __all__ = [

python/paddle/v2/fluid/io.py

@@ -1,7 +1,7 @@
 import os
 import cPickle as pickle
 
-from paddle.v2.framework.framework import Program, Parameter, g_main_program, \
+from paddle.v2.fluid.framework import Program, Parameter, g_main_program, \
     Variable
 
 __all__ = [

python/paddle/v2/fluid/layer_helper.py

@@ -1,9 +1,9 @@
 import copy
 import itertools
 
-from paddle.v2.framework.framework import Variable, g_main_program, \
+from paddle.v2.fluid.framework import Variable, g_main_program, \
     g_startup_program, unique_name, Program
-from paddle.v2.framework.initializer import ConstantInitializer, \
+from paddle.v2.fluid.initializer import ConstantInitializer, \
     UniformInitializer, XavierInitializer
 
 

python/paddle/v2/fluid/layers.py

@@ -1,10 +1,10 @@
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
-import paddle.v2.framework.proto.framework_pb2 as framework_pb2
+import paddle.v2.fluid.proto.framework_pb2 as framework_pb2
-from paddle.v2.framework.framework import OpProtoHolder, Variable, Program, \
+from paddle.v2.fluid.framework import OpProtoHolder, Variable, Program, \
     Operator
-from paddle.v2.framework.initializer import ConstantInitializer, \
+from paddle.v2.fluid.initializer import ConstantInitializer, \
     NormalInitializer
-from paddle.v2.framework.layer_helper import LayerHelper, unique_name
+from paddle.v2.fluid.layer_helper import LayerHelper, unique_name
 import re
 import cStringIO
 
@@ -839,6 +839,23 @@ def batch_norm(input,
     return helper.append_activation(batch_norm_out)
 
 
+def beam_search_decode(ids, scores, main_program=None, startup_program=None):
+    helper = LayerHelper('beam_search_decode', **locals())
+    sentence_ids = helper.create_tmp_variable(dtype=ids.data_type)
+    sentence_scores = helper.create_tmp_variable(dtype=ids.data_type)
+
+    helper.append_op(
+        type="beam_search_decode",
+        inputs={"Ids": ids,
+                "Scores": scores},
+        outputs={
+            "SentenceIds": sentence_ids,
+            "SentenceScores": sentence_scores
+        })
+
+    return sentence_ids, sentence_scores
+
+
 class BlockGuard(object):
     """
     BlockGuard class.

python/paddle/v2/fluid/net_drawer.py

@@ -3,8 +3,8 @@ import json
 import logging
 from collections import defaultdict
 
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
-import paddle.v2.framework.proto.framework_pb2 as framework_pb2
+import paddle.v2.fluid.proto.framework_pb2 as framework_pb2
 
 logger = logging.getLogger(__name__)
 logger.setLevel(logging.INFO)

python/paddle/v2/fluid/nets.py

@@ -1,4 +1,4 @@
-import paddle.v2.framework.layers as layers
+import paddle.v2.fluid.layers as layers
 
 __all__ = ["simple_img_conv_pool", "sequence_conv_pool"]
 

python/paddle/v2/fluid/op.py

@@ -1,5 +1,5 @@
-import paddle.v2.framework.core as core
+import paddle.v2.fluid.core as core
-import paddle.v2.framework.proto.framework_pb2 as framework_pb2
+import paddle.v2.fluid.proto.framework_pb2 as framework_pb2
 
 
 def get_all_op_protos():