diff --git a/RELEASE.cn.md b/RELEASE.cn.md
old mode 100755
new mode 100644
diff --git a/cmake/generic.cmake b/cmake/generic.cmake
index 063a09b63e09e3e947833453c0b07b1631e945cf..555faff499eddc9f153f55fa2769da1238f436b4 100644
--- a/cmake/generic.cmake
+++ b/cmake/generic.cmake
@@ -28,6 +28,105 @@
 # cmake_parse_arguments can help us to achieve this goal.
 # https://cmake.org/cmake/help/v3.0/module/CMakeParseArguments.html
 
+# cc_library parses tensor.cc and figures out that target also depend on tensor.h.
+# cc_library(tensor
+#   SRCS
+#   tensor.cc
+#   DEPS
+#   variant)
+function(cc_library TARGET_NAME)
+  set(options OPTIONAL)
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(cc_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  if (${cc_library_OPTIONAL} STREQUAL "SHARED")
+    add_library(${TARGET_NAME} SHARED ${cc_library_SRCS})
+  else()
+    add_library(${TARGET_NAME} STATIC ${cc_library_SRCS})
+  endif()
+  add_dependencies(${TARGET_NAME} ${cc_library_DEPS} ${external_project_dependencies})
+endfunction(cc_library)
+
+# cc_binary parses tensor.cc and figures out that target also depend on tensor.h.
+# cc_binary(tensor
+#   SRCS
+#   tensor.cc)
+function(cc_binary TARGET_NAME)
+  set(options OPTIONAL)
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(cc_binary "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  add_executable(${TARGET_NAME} ${cc_binary_SRCS})
+  add_dependencies(${TARGET_NAME} ${cc_binary_DEPS} ${external_project_dependencies})
+  target_link_libraries(${TARGET_NAME} ${cc_binary_DEPS})
+endfunction(cc_binary)
+
+# The dependency to target tensor implies that if any of
+# tensor{.h,.cc,_test.cc} is changed, tensor_test need to be re-built.
+# cc_test(tensor_test
+#   SRCS
+#   tensor_test.cc
+#   DEPS
+#   tensor)
+function(cc_test TARGET_NAME)
+  set(options "")
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  add_executable(${TARGET_NAME} ${cc_test_SRCS})
+  add_dependencies(${TARGET_NAME} ${cc_test_DEPS} ${external_project_dependencies})
+  target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} ${GTEST_MAIN_LIBRARIES} ${GTEST_LIBRARIES})
+  add_test(${TARGET_NAME} ${TARGET_NAME})
+endfunction(cc_test)
+
+# Suppose that ops.cu includes global functions that take Tensor as
+# their parameters, so ops depend on tensor. This implies that if
+# any of tensor.{h.cc}, ops.{h,cu} is changed, ops need to be re-built.
+# nv_library(ops
+#   SRCS
+#   ops.cu
+#   DEPS
+#   tensor)
+function(nv_library TARGET_NAME)
+  set(options OPTIONAL)
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(nv_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  if (${nv_library_OPTIONAL} STREQUAL "SHARED")
+    cuda_add_library(${TARGET_NAME} SHARED ${nv_library_SRCS})
+  else()
+    cuda_add_library(${TARGET_NAME} STATIC ${nv_library_SRCS})
+  endif()
+  add_dependencies(${TARGET_NAME} ${nv_library_DEPS} ${external_project_dependencies})
+endfunction(nv_library)
+
+function(nv_binary TARGET_NAME)
+  set(options "")
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(nv_binary "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  cuda_add_executable(${TARGET_NAME} ${nv_binary_SRCS})
+  add_dependencies(${TARGET_NAME} ${nv_binary_DEPS} ${external_project_dependencies})
+  target_link_libraries(${TARGET_NAME} ${nv_binary_DEPS})
+endfunction(nv_binary)
+
+# The dependency to target tensor implies that if any of
+# ops{.h,.cu,_test.cu} is changed, ops_test need to be re-built.
+# nv_test(ops_test
+#   SRCS
+#   ops_test.cu
+#   DEPS
+#   ops)
+function(nv_test TARGET_NAME)
+  set(options "")
+  set(oneValueArgs "")
+  set(multiValueArgs SRCS DEPS)
+  cmake_parse_arguments(nv_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  cuda_add_executable(${TARGET_NAME} ${nv_test_SRCS})
+  add_dependencies(${TARGET_NAME} ${nv_test_DEPS} ${external_project_dependencies})
+  target_link_libraries(${TARGET_NAME} ${nv_test_DEPS} ${GTEST_MAIN_LIBRARIES} ${GTEST_LIBRARIES})
+  add_test(${TARGET_NAME} ${TARGET_NAME})
+endfunction(nv_test)
 
 set(GOPATH "${CMAKE_CURRENT_BINARY_DIR}/go")
 file(MAKE_DIRECTORY ${GOPATH})
diff --git a/demo/semantic_role_labeling/api_train_v2.py b/demo/semantic_role_labeling/api_train_v2.py
index 036cad4b0a32357bb42580ef577a1eba558be8fe..3af636aef5879b43641d55bd7c9b0b8a1242ff8b 100644
--- a/demo/semantic_role_labeling/api_train_v2.py
+++ b/demo/semantic_role_labeling/api_train_v2.py
@@ -1,26 +1,33 @@
-import sys
 import math
 import numpy as np
-import paddle.v2 as paddle
+import gzip
+import logging
 import paddle.v2.dataset.conll05 as conll05
+import paddle.v2.evaluator as evaluator
+import paddle.v2 as paddle
 
+logger = logging.getLogger('paddle')
 
-def db_lstm():
-    word_dict, verb_dict, label_dict = conll05.get_dict()
-    word_dict_len = len(word_dict)
-    label_dict_len = len(label_dict)
-    pred_len = len(verb_dict)
+word_dict, verb_dict, label_dict = conll05.get_dict()
+word_dict_len = len(word_dict)
+label_dict_len = len(label_dict)
+pred_len = len(verb_dict)
 
-    mark_dict_len = 2
-    word_dim = 32
-    mark_dim = 5
-    hidden_dim = 512
-    depth = 8
+mark_dict_len = 2
+word_dim = 32
+mark_dim = 5
+hidden_dim = 512
+depth = 8
+default_std = 1 / math.sqrt(hidden_dim) / 3.0
+mix_hidden_lr = 1e-3
 
-    #8 features
-    def d_type(size):
-        return paddle.data_type.integer_value_sequence(size)
 
+def d_type(size):
+    return paddle.data_type.integer_value_sequence(size)
+
+
+def db_lstm():
+    #8 features
     word = paddle.layer.data(name='word_data', type=d_type(word_dict_len))
     predicate = paddle.layer.data(name='verb_data', type=d_type(pred_len))
 
@@ -31,11 +38,7 @@ def db_lstm():
     ctx_p2 = paddle.layer.data(name='ctx_p2_data', type=d_type(word_dict_len))
     mark = paddle.layer.data(name='mark_data', type=d_type(mark_dict_len))
 
-    target = paddle.layer.data(name='target', type=d_type(label_dict_len))
-
-    default_std = 1 / math.sqrt(hidden_dim) / 3.0
-
-    emb_para = paddle.attr.Param(name='emb', initial_std=0., learning_rate=0.)
+    emb_para = paddle.attr.Param(name='emb', initial_std=0., is_static=True)
     std_0 = paddle.attr.Param(initial_std=0.)
     std_default = paddle.attr.Param(initial_std=default_std)
 
@@ -63,7 +66,6 @@ def db_lstm():
                 input=emb, param_attr=std_default) for emb in emb_layers
         ])
 
-    mix_hidden_lr = 1e-3
     lstm_para_attr = paddle.attr.Param(initial_std=0.0, learning_rate=1.0)
     hidden_para_attr = paddle.attr.Param(
         initial_std=default_std, learning_rate=mix_hidden_lr)
@@ -111,6 +113,21 @@ def db_lstm():
                 input=input_tmp[1], param_attr=lstm_para_attr)
         ], )
 
+    return feature_out
+
+
+def load_parameter(file_name, h, w):
+    with open(file_name, 'rb') as f:
+        f.read(16)  # skip header.
+        return np.fromfile(f, dtype=np.float32).reshape(h, w)
+
+
+def train():
+    paddle.init(use_gpu=False, trainer_count=1)
+
+    # define network topology
+    feature_out = db_lstm()
+    target = paddle.layer.data(name='target', type=d_type(label_dict_len))
     crf_cost = paddle.layer.crf(size=label_dict_len,
                                 input=feature_out,
                                 label=target,
@@ -120,29 +137,15 @@ def db_lstm():
                                     learning_rate=mix_hidden_lr))
 
     crf_dec = paddle.layer.crf_decoding(
-        name='crf_dec_l',
         size=label_dict_len,
         input=feature_out,
         label=target,
         param_attr=paddle.attr.Param(name='crfw'))
-
-    return crf_cost, crf_dec
-
-
-def load_parameter(file_name, h, w):
-    with open(file_name, 'rb') as f:
-        f.read(16)  # skip header.
-        return np.fromfile(f, dtype=np.float32).reshape(h, w)
-
-
-def main():
-    paddle.init(use_gpu=False, trainer_count=1)
-
-    # define network topology
-    crf_cost, crf_dec = db_lstm()
+    evaluator.sum(input=crf_dec)
 
     # create parameters
-    parameters = paddle.parameters.create([crf_cost, crf_dec])
+    parameters = paddle.parameters.create(crf_cost)
+    parameters.set('emb', load_parameter(conll05.get_embedding(), 44068, 32))
 
     # create optimizer
     optimizer = paddle.optimizer.Momentum(
@@ -152,18 +155,12 @@ def main():
         model_average=paddle.optimizer.ModelAverage(
             average_window=0.5, max_average_window=10000), )
 
-    def event_handler(event):
-        if isinstance(event, paddle.event.EndIteration):
-            if event.batch_id % 100 == 0:
-                print "Pass %d, Batch %d, Cost %f, %s" % (
-                    event.pass_id, event.batch_id, event.cost, event.metrics)
-
     trainer = paddle.trainer.SGD(cost=crf_cost,
                                  parameters=parameters,
-                                 update_equation=optimizer)
-    parameters.set('emb', load_parameter(conll05.get_embedding(), 44068, 32))
+                                 update_equation=optimizer,
+                                 extra_layers=crf_dec)
 
-    trn_reader = paddle.batch(
+    reader = paddle.batch(
         paddle.reader.shuffle(
             conll05.test(), buf_size=8192), batch_size=10)
 
@@ -179,12 +176,102 @@ def main():
         'target': 8
     }
 
+    def event_handler(event):
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 100 == 0:
+                logger.info("Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics))
+            if event.batch_id and event.batch_id % 1000 == 0:
+                result = trainer.test(reader=reader, feeding=feeding)
+                logger.info("\nTest with Pass %d, Batch %d, %s" %
+                            (event.pass_id, event.batch_id, result.metrics))
+
+        if isinstance(event, paddle.event.EndPass):
+            # save parameters
+            with gzip.open('params_pass_%d.tar.gz' % event.pass_id, 'w') as f:
+                parameters.to_tar(f)
+
+            result = trainer.test(reader=reader, feeding=feeding)
+            logger.info("\nTest with Pass %d, %s" %
+                        (event.pass_id, result.metrics))
+
     trainer.train(
-        reader=trn_reader,
+        reader=reader,
         event_handler=event_handler,
-        num_passes=10000,
+        num_passes=10,
         feeding=feeding)
 
 
+def infer_a_batch(inferer, test_data, word_dict, pred_dict, label_dict):
+    probs = inferer.infer(input=test_data, field='id')
+    assert len(probs) == sum(len(x[0]) for x in test_data)
+
+    for idx, test_sample in enumerate(test_data):
+        start_id = 0
+        pred_str = "%s\t" % (pred_dict[test_sample[6][0]])
+
+        for w, tag in zip(test_sample[0],
+                          probs[start_id:start_id + len(test_sample[0])]):
+            pred_str += "%s[%s] " % (word_dict[w], label_dict[tag])
+        print(pred_str.strip())
+        start_id += len(test_sample[0])
+
+
+def infer():
+    label_dict_reverse = dict((value, key)
+                              for key, value in label_dict.iteritems())
+    word_dict_reverse = dict((value, key)
+                             for key, value in word_dict.iteritems())
+    pred_dict_reverse = dict((value, key)
+                             for key, value in verb_dict.iteritems())
+
+    test_creator = paddle.dataset.conll05.test()
+
+    paddle.init(use_gpu=False, trainer_count=1)
+
+    # define network topology
+    feature_out = db_lstm()
+    predict = paddle.layer.crf_decoding(
+        size=label_dict_len,
+        input=feature_out,
+        param_attr=paddle.attr.Param(name='crfw'))
+
+    test_pass = 0
+    with gzip.open('params_pass_%d.tar.gz' % (test_pass)) as f:
+        parameters = paddle.parameters.Parameters.from_tar(f)
+        inferer = paddle.inference.Inference(
+            output_layer=predict, parameters=parameters)
+
+        # prepare test data
+        test_data = []
+        test_batch_size = 50
+
+        for idx, item in enumerate(test_creator()):
+            test_data.append(item[0:8])
+
+            if idx and (not idx % test_batch_size):
+                infer_a_batch(
+                    inferer,
+                    test_data,
+                    word_dict_reverse,
+                    pred_dict_reverse,
+                    label_dict_reverse, )
+                test_data = []
+        infer_a_batch(
+            inferer,
+            test_data,
+            word_dict_reverse,
+            pred_dict_reverse,
+            label_dict_reverse, )
+        test_data = []
+
+
+def main(is_inferring=False):
+    if is_inferring:
+        infer()
+    else:
+        train()
+
+
 if __name__ == '__main__':
-    main()
+    main(is_inferring=False)
diff --git a/paddle/majel/CMakeLists.txt b/paddle/majel/CMakeLists.txt
index d4977df1185b4c13b7c67e24a80fa479e23d46d4..d4bce38906e9326992f6a44ac5cf25309063806a 100644
--- a/paddle/majel/CMakeLists.txt
+++ b/paddle/majel/CMakeLists.txt
@@ -1,43 +1,4 @@
-cmake_minimum_required(VERSION 3.0)
-
-if(${CMAKE_CURRENT_SOURCE_DIR} STREQUAL ${CMAKE_SOURCE_DIR})
-    # find #include <majel/xx.h>
-    get_filename_component(PARENT_DIR ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY)
-    include_directories(${PARENT_DIR})
-
-    # find cmake directory modules
-    get_filename_component(PARENT_DIR ${PARENT_DIR} DIRECTORY)
-    set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${PARENT_DIR}/cmake")
-
-    # enable boost
-    find_package(Boost REQUIRED)
-    if(NOT Boost_FOUND)
-        message(FATAL "Cannot find Boost library.")
-    endif()
-    include_directories(${Boost_INCLUDE_DIRS})
-
-    # enable c++11
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
-
-    # enable gtest
-    set(THIRD_PARTY_PATH ${CMAKE_CURRENT_SOURCE_DIR}/third_party)
-    set(WITH_TESTING ON)
-    include(external/gtest)
-else()
-    message("-- Found gtest (include: ${GTEST_INCLUDE_DIR}, library: ${GTEST_LIBRARIES})")
-endif()
-
-########################### Build Majel #############################
-set(MAJEL_CXX_FILES place.cc)
-set(MAJEL_CUDA_FILES "")
-
-if(CUDA_FOUND)
-    cuda_add_library(majel ${MAJEL_CUDA_FILES} ${MAJEL_CXX_FILES})
-else()
-    add_library(majel ${MAJEL_CXX_FILES})
-endif()
-add_dependencies(majel ${external_project_dependencies})
-#####################################################################
+cc_library(majel SRCS place.cc)
 
 if(WITH_TESTING)
     add_subdirectory(test)
diff --git a/paddle/majel/test/CMakeLists.txt b/paddle/majel/test/CMakeLists.txt
index 76327fdd70c3f1763abda1f38c137dbaf27fba30..68f9059874aed8843da1fc598c7d2e57e9b8bbfe 100644
--- a/paddle/majel/test/CMakeLists.txt
+++ b/paddle/majel/test/CMakeLists.txt
@@ -1,10 +1,7 @@
-file(GLOB_RECURSE ALL_TEST_FILES RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "*.cc")
+cc_test(place_test
+    SRCS place_test.cc
+    DEPS majel)
 
-add_executable(majel_tests ${ALL_TEST_FILES})
-add_dependencies(majel_tests majel)
-target_link_libraries(majel_tests     
-                      ${GTEST_LIBRARIES}
-                      ${GTEST_MAIN_LIBRARIES}
-                      majel
-                     )
-add_test(majel_tests majel_tests)
+if(WITH_GPU)
+    nv_test(cuda_test SRCS cuda_test.cu)
+endif()
diff --git a/paddle/majel/test/cuda_test.cu b/paddle/majel/test/cuda_test.cu
new file mode 100644
index 0000000000000000000000000000000000000000..4067dda2f19f7661722d8a14a27c7b32ed6afc92
--- /dev/null
+++ b/paddle/majel/test/cuda_test.cu
@@ -0,0 +1,59 @@
+#include <cuda_runtime.h>
+#include <stdio.h>
+#include "gtest/gtest.h"
+
+#define CHECK_ERR(x)                 \
+  if (x != cudaSuccess) {            \
+    fprintf(stderr,                  \
+            "%s in %s at line %d\n", \
+            cudaGetErrorString(err), \
+            __FILE__,                \
+            __LINE__);               \
+    exit(-1);                        \
+  }
+
+__global__ void vecAdd(float *d_A, float *d_B, float *d_C, int n) {
+  int i = blockDim.x * blockIdx.x + threadIdx.x;
+  if (i < n) {
+    d_C[i] = d_A[i] + d_B[i];
+  }
+}
+
+TEST(Cuda, Equality) {
+  int n = 10;
+  // Memory allocation for h_A, h_B and h_C (in the host)
+  float h_A[10] = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 0.0};
+  float h_B[10] = {0.0, 9.0, 8.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0};
+  float h_C[10];
+  float *d_A, *d_B, *d_C;
+  cudaError_t err;
+  // Memory allocation for d_A, d_B and d_C (in the device)
+  err = cudaMalloc((void **)&d_A, sizeof(float) * n);
+  CHECK_ERR(err);
+
+  err = cudaMalloc((void **)&d_B, sizeof(float) * n);
+  CHECK_ERR(err);
+
+  err = cudaMalloc((void **)&d_C, sizeof(float) * n);
+  CHECK_ERR(err);
+
+  // Copying memory to device
+  err = cudaMemcpy(d_A, h_A, sizeof(float) * n, cudaMemcpyHostToDevice);
+  CHECK_ERR(err);
+
+  err = cudaMemcpy(d_B, h_B, sizeof(float) * n, cudaMemcpyHostToDevice);
+  CHECK_ERR(err);
+
+  // Calling the kernel
+  vecAdd<<<ceil(n / 256.0), 256>>>(d_A, d_B, d_C, n);
+
+  // Copying results back to host
+  err = cudaMemcpy(h_C, d_C, sizeof(float) * n, cudaMemcpyDeviceToHost);
+  CHECK_ERR(err);
+
+  EXPECT_EQ(h_C[0], 1.0);
+  for (int i = 1; i < n - 1; ++i) {
+    EXPECT_EQ(h_C[i], 11.0);
+  }
+  EXPECT_EQ(h_C[9], 1.0);
+}
diff --git a/paddle/scripts/travis/docs.sh b/paddle/scripts/travis/docs.sh
index 67b89adb4ddb7bb93cb776d64711078cb11a2784..c784293695bf134b5e990639778b6e84ba45d00d 100755
--- a/paddle/scripts/travis/docs.sh
+++ b/paddle/scripts/travis/docs.sh
@@ -60,7 +60,6 @@ function deploy_docs() {
 
 deploy_docs "master" "." 
 deploy_docs "develop" "./develop/"
-deploy_docs "release/0.10.0" "./release/0.10.0/"
 
 # Check is there anything changed.
 set +e