Merge pull request #1013 from wangjiawei04/v0.4.1

cherry-pick 1008 1010

cmake/paddlepaddle.cmake

@@ -32,19 +32,23 @@ message( "WITH_GPU = ${WITH_GPU}")
 # latest: latest develop build
 # version number like 1.5.2
 SET(PADDLE_VERSION "2.0.0")
-if(CUDA_VERSION EQUAL 11.0)
-    set(CUDA_SUFFIX "cuda11-cudnn8-avx-mkl")
-    set(WITH_TRT ON)
-elseif(CUDA_VERSION EQUAL 10.2)
-    set(CUDA_SUFFIX "cuda10.2-cudnn8-avx-mkl")
-    set(WITH_TRT ON)
-elseif(CUDA_VERSION EQUAL 10.1)
-    set(CUDA_SUFFIX "cuda10.1-cudnn7-avx-mkl")
-    set(WITH_TRT ON)
-elseif(CUDA_VERSION EQUAL 10.0)
-    set(CUDA_SUFFIX "cuda10-cudnn7-avx-mkl")
-elseif(CUDA_VERSION EQUAL 9.0)
-    set(CUDA_SUFFIX "cuda9-cudnn7-avx-mkl")
+if (WITH_GPU)
+    if(CUDA_VERSION EQUAL 11.0)
+        set(CUDA_SUFFIX "cuda11-cudnn8-avx-mkl")
+        set(WITH_TRT ON)
+    elseif(CUDA_VERSION EQUAL 10.2)
+        set(CUDA_SUFFIX "cuda10.2-cudnn8-avx-mkl")
+        set(WITH_TRT ON)
+    elseif(CUDA_VERSION EQUAL 10.1)
+        set(CUDA_SUFFIX "cuda10.1-cudnn7-avx-mkl")
+        set(WITH_TRT ON)
+    elseif(CUDA_VERSION EQUAL 10.0)
+        set(CUDA_SUFFIX "cuda10-cudnn7-avx-mkl")
+    elseif(CUDA_VERSION EQUAL 9.0)
+        set(CUDA_SUFFIX "cuda9-cudnn7-avx-mkl")
+    endif()
+else()
+    set(WITH_TRT OFF)
 endif()  
 
 if (WITH_GPU)

core/cube/cube-builder/CMakeLists.txt

@@ -22,7 +22,7 @@ include_directories(SYSTEM ${CMAKE_CURRENT_BINARY_DIR}/../)
 
 add_executable(cube-builder src/main.cpp include/cube-builder/util.h src/util.cpp src/builder_job.cpp include/cube-builder/builder_job.h include/cube-builder/define.h src/seqfile_reader.cpp include/cube-builder/seqfile_reader.h include/cube-builder/raw_reader.h include/cube-builder/vtext.h src/crovl_builder_increment.cpp include/cube-builder/crovl_builder_increment.h src/curl_simple.cpp include/cube-builder/curl_simple.h)
 
-add_dependencies(cube-builder jsoncpp)
+add_dependencies(cube-builder jsoncpp boost)
 
 set(DYNAMIC_LIB
     gflags

python/pipeline/operator.py

@@ -709,11 +709,39 @@ class Op(object):
             # combine samples to batch
             one_input = preped_data_dict[data_ids[0]]
             feed_batch = []
+            feed_dict = {}
             input_offset = None
+            cur_offset = 0
+            input_offset_dict = {}
+
             if isinstance(one_input, dict):
                 # sample input
-                feed_batch = [preped_data_dict[data_id] for data_id in data_ids]
-                input_offset = list(range(len(data_ids) + 1))
+                if len(data_ids) == 1:
+                    feed_batch = [
+                        preped_data_dict[data_id] for data_id in data_ids
+                    ]
+                else:
+                    for data_id in data_ids:
+                        for key, val in preped_data_dict[data_id].items():
+                            has_val = feed_dict.get(key)
+                            if has_val is None:
+                                feed_dict[key] = val
+                                continue
+                            # merge 2 np.arrray
+                            if isinstance(val, np.ndarray):
+                                feed_dict[key] = np.append(
+                                    feed_dict[key], val, axis=0)
+                    feed_batch.append(feed_dict)
+
+                for data_id in data_ids:
+                    start = cur_offset
+                    for key, val in preped_data_dict[data_id].items():
+                        if isinstance(val, (list, np.ndarray)):
+                            cur_offset += len(val)
+                        else:
+                            cur_offset += 1
+                        break
+                    input_offset_dict[data_id] = [start, cur_offset]
             elif isinstance(one_input, list):
                 # batch input
                 input_offset = [0]
@@ -796,8 +824,10 @@ class Op(object):
                 for name in var_names:
                     lod_offset_name = "{}.lod".format(name)
                     if lod_offset_name in var_names:
-                        _LOGGER.debug("(log_id={}) {} {} is LodTensor".format(
-                            typical_logid, op_info_prefix, name))
+                        _LOGGER.debug(
+                            "(log_id={}) {} {} is LodTensor. lod_offset_name:{}".
+                            format(typical_logid, op_info_prefix, name,
+                                   lod_offset_name))
                         lod_var_names.add(name)
                         lod_offset_names.add(lod_offset_name)
 
@@ -812,8 +842,8 @@ class Op(object):
                         lod_offset_name = "{}.lod".format(name)
                         lod_offset = midped_batch[lod_offset_name]
                         for idx, data_id in enumerate(data_ids):
-                            data_offset_left = input_offset[idx]
-                            data_offset_right = input_offset[idx + 1]
+                            data_offset_left = input_offset_dict[data_id][0]
+                            data_offset_right = input_offset_dict[data_id][1]
                             lod_offset_left = lod_offset[data_offset_left]
                             lod_offset_right = lod_offset[data_offset_right]
                             midped_data_dict[data_id][name] = value[
@@ -823,8 +853,8 @@ class Op(object):
                     else:
                         # normal tensor
                         for idx, data_id in enumerate(data_ids):
-                            left = input_offset[idx]
-                            right = input_offset[idx + 1]
+                            left = input_offset_dict[data_id][0]
+                            right = input_offset_dict[data_id][1]
                             midped_data_dict[data_id][name] = value[left:right]
         else:
             midped_data_dict = preped_data_dict

tools/Dockerfile.ci

-FROM centos:7.3.1611
+FROM hub.baidubce.com/paddlepaddle/serving:latest-cuda10.2-cudnn8-devel
 
-RUN yum -y install wget >/dev/null \
-    && yum -y install gcc gcc-c++ make glibc-static which >/dev/null \
-    && yum -y install git openssl-devel curl-devel bzip2-devel python-devel >/dev/null \
-    && yum -y install libSM-1.2.2-2.el7.x86_64 --setopt=protected_multilib=false \
-    && yum -y install libXrender-0.9.10-1.el7.x86_64 --setopt=protected_multilib=false \
-    && yum -y install libXext-1.3.3-3.el7.x86_64 --setopt=protected_multilib=false
-    
-RUN wget https://cmake.org/files/v3.2/cmake-3.2.0-Linux-x86_64.tar.gz >/dev/null \
-    && tar xzf cmake-3.2.0-Linux-x86_64.tar.gz \
-    && mv cmake-3.2.0-Linux-x86_64 /usr/local/cmake3.2.0 \
-    && echo 'export PATH=/usr/local/cmake3.2.0/bin:$PATH' >> /root/.bashrc \
-    && rm cmake-3.2.0-Linux-x86_64.tar.gz
-
-RUN wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
-    && tar xzf go1.14.linux-amd64.tar.gz \
-    && mv go /usr/local/go \
-    && echo 'export GOROOT=/usr/local/go' >> /root/.bashrc \
-    && echo 'export PATH=/usr/local/go/bin:$PATH' >> /root/.bashrc \
-    && rm go1.14.linux-amd64.tar.gz
-    
-RUN yum -y install python-devel sqlite-devel >/dev/null \
-    && curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
-    && python get-pip.py >/dev/null \
-    && rm get-pip.py \
-    && pip install requests 
-
-RUN wget http://nixos.org/releases/patchelf/patchelf-0.10/patchelf-0.10.tar.bz2 \
-    && yum -y install bzip2 >/dev/null \
-    && tar -jxf patchelf-0.10.tar.bz2 \
-    && cd patchelf-0.10 \
-    && ./configure --prefix=/usr \
-    && make >/dev/null && make install >/dev/null \
-    && cd .. \
-    && rm -rf patchelf-0.10*
-
-RUN yum install -y python3 python3-devel \
-    && pip3 install requests
-
-RUN wget https://github.com/protocolbuffers/protobuf/releases/download/v3.11.2/protobuf-all-3.11.2.tar.gz && \
-    tar zxf protobuf-all-3.11.2.tar.gz && \
-    cd protobuf-3.11.2 && \
-    ./configure && make -j4 && make install && \
-    make clean && \
-    cd .. && rm -rf protobuf-*
-
-RUN yum -y update >/dev/null \
-    && yum -y install dnf >/dev/null \
-    && yum -y install dnf-plugins-core >/dev/null \
-    && dnf copr enable alonid/llvm-3.8.0 -y \
-    && dnf install llvm-3.8.0 clang-3.8.0 compiler-rt-3.8.0 -y \
-    && echo 'export PATH=/opt/llvm-3.8.0/bin:$PATH' >> /root/.bashrc
-
-RUN yum install -y java \
-    && wget http://repos.fedorapeople.org/repos/dchen/apache-maven/epel-apache-maven.repo -O /etc/yum.repos.d/epel-apache-maven.repo \
-    && yum install -y apache-maven
-
-RUN yum install -y lsof