diff --git a/BUILD.md b/BUILD.md
index b97ef6fe43..f2a41d20ff 100644
--- a/BUILD.md
+++ b/BUILD.md
@@ -420,14 +420,14 @@ index 7dfa97c..6d99e71 100644
 #### Build Instructions
 ##### Windows
 ```
-.\build.bat --use_tvm --use_llvm --llvm_path=\llvm\install\path\lib\cmake\llvm --use_mklml --use_nuphar --build_shared_lib --build_csharp --enable_pybind --config=Release
+.\build.bat --llvm_path=\llvm\install\path\lib\cmake\llvm --use_mklml --use_nuphar --build_shared_lib --build_csharp --enable_pybind --config=Release
 ```
 
 * These instructions build the release flavor. The Debug build of LLVM would be needed to build with the Debug flavor of ONNX Runtime.
 
 ##### Linux:
 ```
-./build.sh --use_tvm --use_llvm --llvm_path=/llvm/install/path/lib/cmake/llvm --use_mklml --use_nuphar --build_shared_lib --build_csharp --enable_pybind --config=Release
+./build.sh --llvm_path=/llvm/install/path/lib/cmake/llvm --use_mklml --use_nuphar --build_shared_lib --build_csharp --enable_pybind --config=Release
 ```
 
 Dockerfile instructions are available [here](./dockerfiles#nuphar).
diff --git a/cgmanifests/cgmanifest.json b/cgmanifests/cgmanifest.json
index c0cc258dc3..6543b2a554 100644
--- a/cgmanifests/cgmanifest.json
+++ b/cgmanifests/cgmanifest.json
@@ -337,6 +337,16 @@
             },
             "comments": "middleware between IB verbs and OpenMPI used by onnxruntime training image"
          }
+      },
+      {
+         "component": {
+            "type": "git",
+            "git": {
+               "commitHash": "63d1e08e64e7e09408eb63cd8dd7c65ad766f277",
+               "repositoryUrl": "https://github.com/nodejs/node"
+            },
+            "comments": "For Nodejs binding"
+         }
       }
    ],
    "Version": 1
diff --git a/cmake/CMakeLists.txt b/cmake/CMakeLists.txt
index f0a07dfb93..f4c3d7d297 100644
--- a/cmake/CMakeLists.txt
+++ b/cmake/CMakeLists.txt
@@ -5,6 +5,8 @@
 # Minimum CMake required
 cmake_minimum_required(VERSION 3.13)
 cmake_policy(SET CMP0069 NEW)
+set(CMAKE_POLICY_DEFAULT_CMP0069 NEW)
+
 if (${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.15")
   cmake_policy(SET CMP0092 NEW)
 endif()
@@ -178,7 +180,7 @@ if(onnxruntime_USE_OPENMP)
     include_directories(${OpenMP_CXX_INCLUDE_DIR})
     # MKLML and NGraph depend on their own OpenMP library that may be different with the compiler's.
     # Disable the options to build mklml/NGraph and OpenMP together.
-    if(onnxruntime_USE_MKLML)
+    if((WIN32 OR APPLE) AND onnxruntime_USE_MKLML)
       message(FATAL_ERROR "Please use only one of onnxruntime_USE_MKLML, onnxruntime_USE_OPENMP")
     endif()
     if(onnxruntime_USE_NGRAPH)
@@ -759,19 +761,17 @@ if (onnxruntime_USE_MKLML)
   if (WIN32 OR APPLE)
     list(APPEND onnxruntime_EXTERNAL_LIBRARIES mklml)
   else()
-    list(APPEND onnxruntime_EXTERNAL_LIBRARIES mklml_intel)
+    if(onnxruntime_USE_OPENMP)
+      list(APPEND onnxruntime_EXTERNAL_LIBRARIES mklml_gnu)
+    else()
+      list(APPEND onnxruntime_EXTERNAL_LIBRARIES mklml_intel)
+    endif()
   endif()
   list(APPEND onnxruntime_EXTERNAL_DEPENDENCIES project_mklml)
   include_directories(${MKLML_INCLUDE_DIR})
   link_directories(${MKLML_LIB_DIR})
 endif()
 
-if (onnxruntime_USE_DNNL)
-  list(APPEND onnxruntime_EXTERNAL_LIBRARIES dnnl)
-  list(APPEND onnxruntime_EXTERNAL_DEPENDENCIES project_dnnl)
-  link_directories(${DNNL_LIB_DIR})
-endif()
-
 if (onnxruntime_USE_NGRAPH)
   #if (onnxruntime_USE_OPENMP)
   #  message(FATAL_ERROR "Please set onnxruntime_USE_OPENMP=OFF for nGraph execution provider.")
@@ -941,7 +941,7 @@ if (onnxruntime_USE_TVM)
     # wd4456: declaration of X hides previous local declaration
     set(DISABLED_WARNINGS_FOR_TVM "/wd4100" "/wd4244" "/wd4251" "/wd4267" "/wd4275" "/wd4389" "/wd4456")
   else()
-    set(DISABLED_WARNINGS_FOR_TVM "-Wno-error=ignored-qualifiers")
+    set(DISABLED_WARNINGS_FOR_TVM "-Wno-error=ignored-qualifiers" "-Wno-unused-parameter" "-Wno-ignored-qualifiers")
     if(HAS_UNUSED_PARAMETER)
       list(APPEND DISABLED_WARNINGS_FOR_TVM "-Wno-error=unused-parameter")
     endif()
@@ -1166,14 +1166,8 @@ endif()
 # built; hence the ordering
 if (onnxruntime_BUILD_UNIT_TESTS)
   if (onnxruntime_ENABLE_PYTHON)
-    if(UNIX)
-      set(CMAKE_SKIP_BUILD_RPATH ON)
-    endif()
     include(onnxruntime_python.cmake)
   endif()
-  # we need to make sure this is turned off since it
-  # turned ON by the previous step when building a shared lib
-  set(CMAKE_SKIP_BUILD_RPATH OFF)
   include(onnxruntime_unittests.cmake)
 endif()
 
diff --git a/cmake/external/dnnl.cmake b/cmake/external/dnnl.cmake
index f67ecbf49c..8975537b0d 100644
--- a/cmake/external/dnnl.cmake
+++ b/cmake/external/dnnl.cmake
@@ -23,14 +23,22 @@ else()
   if (APPLE)
     set(DNNL_SHARED_LIB libdnnl.1.dylib)
     set(MKLML_OS_VERSION_STR "mac")
+    if(onnxruntime_USE_MKLML)
+      set(MKLML_SHARED_LIB libmklml.dylib)
+      set(IOMP5MD_SHARED_LIB libiomp5.dylib)
+    endif()
   else()
     set(DNNL_SHARED_LIB libdnnl.so.1)
     set(MKLML_OS_VERSION_STR "lnx")
-  endif()
-  if(onnxruntime_USE_MKLML)
-    set(MKLML_SHARED_LIB libmklml_intel.so)
-    set(IOMP5MD_SHARED_LIB libiomp5.so)
-  endif()
+    if(onnxruntime_USE_MKLML)
+      if(onnxruntime_USE_OPENMP)
+        set(MKLML_SHARED_LIB libmklml_gnu.so)
+      else()
+        set(MKLML_SHARED_LIB libmklml_intel.so)
+        set(IOMP5MD_SHARED_LIB libiomp5.so)
+      endif()
+    endif()
+  endif()  
 endif()
 
 if (onnxruntime_USE_MKLML)
diff --git a/cmake/onnxruntime_providers.cmake b/cmake/onnxruntime_providers.cmake
index 9172f8851a..659f4057da 100644
--- a/cmake/onnxruntime_providers.cmake
+++ b/cmake/onnxruntime_providers.cmake
@@ -35,10 +35,6 @@ file(GLOB onnxruntime_providers_common_srcs CONFIGURE_DEPENDS
   "${ONNXRUNTIME_ROOT}/core/providers/*.cc"
 )
 
-if(onnxruntime_USE_DNNL)
-  set(PROVIDERS_DNNL onnxruntime_providers_dnnl)
-  list(APPEND ONNXRUNTIME_PROVIDER_NAMES dnnl)
-endif()
 if(onnxruntime_USE_NGRAPH)
   set(PROVIDERS_NGRAPH onnxruntime_providers_ngraph)
   list(APPEND ONNXRUNTIME_PROVIDER_NAMES ngraph)
@@ -314,6 +310,8 @@ if (onnxruntime_USE_TENSORRT OR onnxruntime_USE_DNNL)
 endif()
 
 if (onnxruntime_USE_DNNL)
+  list(APPEND ONNXRUNTIME_PROVIDER_NAMES dnnl)
+
   file(GLOB_RECURSE onnxruntime_providers_dnnl_cc_srcs CONFIGURE_DEPENDS
     "${ONNXRUNTIME_ROOT}/core/providers/dnnl/*.h"
     "${ONNXRUNTIME_ROOT}/core/providers/dnnl/*.cc"
@@ -323,8 +321,9 @@ if (onnxruntime_USE_DNNL)
 
   source_group(TREE ${ONNXRUNTIME_ROOT}/core FILES ${onnxruntime_providers_dnnl_cc_srcs})
   add_library(onnxruntime_providers_dnnl SHARED ${onnxruntime_providers_dnnl_cc_srcs})
+  target_link_directories(onnxruntime_providers_dnnl PRIVATE ${DNNL_LIB_DIR})
   onnxruntime_add_include_to_target(onnxruntime_providers_dnnl onnxruntime_common onnx) # onnx needed for stl_backports.h
-  add_dependencies(onnxruntime_providers_dnnl onnxruntime_providers_shared ${onnxruntime_EXTERNAL_DEPENDENCIES})
+  add_dependencies(onnxruntime_providers_dnnl onnxruntime_providers_shared project_dnnl ${onnxruntime_EXTERNAL_DEPENDENCIES})
   target_include_directories(onnxruntime_providers_dnnl PRIVATE ${ONNXRUNTIME_ROOT} ${eigen_INCLUDE_DIRS} ${DNNL_INCLUDE_DIR})
   target_link_libraries(onnxruntime_providers_dnnl PRIVATE dnnl onnxruntime_providers_shared)
   install(DIRECTORY ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/providers/dnnl  DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core/providers)
diff --git a/cmake/onnxruntime_unittests.cmake b/cmake/onnxruntime_unittests.cmake
index fcc40dd950..8726558543 100644
--- a/cmake/onnxruntime_unittests.cmake
+++ b/cmake/onnxruntime_unittests.cmake
@@ -560,22 +560,23 @@ add_custom_command(
   COMMAND ${CMAKE_COMMAND} -E copy_directory
   ${TEST_DATA_SRC}
   ${TEST_DATA_DES})
-if(WIN32)
-  if (onnxruntime_USE_DNNL)
-    list(APPEND onnx_test_libs dnnl)
-    add_custom_command(
-      TARGET ${test_data_target} POST_BUILD
-      COMMAND ${CMAKE_COMMAND} -E copy ${DNNL_DLL_PATH} $<TARGET_FILE_DIR:${test_data_target}>
-      )
-  endif()
-  if (onnxruntime_USE_MKLML)
-    add_custom_command(
-      TARGET ${test_data_target} POST_BUILD
-      COMMAND ${CMAKE_COMMAND} -E copy
-      ${MKLML_LIB_DIR}/${MKLML_SHARED_LIB} ${MKLML_LIB_DIR}/${IOMP5MD_SHARED_LIB}
-      $<TARGET_FILE_DIR:${test_data_target}>
+
+if (onnxruntime_USE_DNNL)
+  list(APPEND onnx_test_libs dnnl)
+  add_custom_command(
+    TARGET ${test_data_target} POST_BUILD
+    COMMAND ${CMAKE_COMMAND} -E copy ${DNNL_DLL_PATH} $<TARGET_FILE_DIR:${test_data_target}>
     )
-  endif()
+endif()
+if (onnxruntime_USE_MKLML)
+  add_custom_command(
+    TARGET ${test_data_target} POST_BUILD
+    COMMAND ${CMAKE_COMMAND} -E copy
+    ${MKLML_LIB_DIR}/${MKLML_SHARED_LIB} ${MKLML_LIB_DIR}/${IOMP5MD_SHARED_LIB}
+    $<TARGET_FILE_DIR:${test_data_target}>
+  )
+endif()
+if(WIN32)
   if (onnxruntime_USE_NGRAPH)
     add_custom_command(
       TARGET ${test_data_target} POST_BUILD
diff --git a/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest-docker.sh b/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest-docker.sh
index 4dc4a05a5e..0f18caf15e 100755
--- a/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest-docker.sh
+++ b/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest-docker.sh
@@ -22,26 +22,8 @@ OldDir=$(pwd)
 
 cd $SOURCE_ROOT/tools/ci_build/github/linux/docker
 
-if [ $UseCentos7 = "false" ]; then
-  echo "Image used for testing is onnxruntime-$IMAGE"
-  if [ $Arch = "x86" ]; then
-    sudo docker build -t "onnxruntime-$IMAGE" --build-arg OS_VERSION=16.04 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=${PYTHON_VER} -f Dockerfile.ubuntu_x86 .
-  else
-    sudo docker build -t "onnxruntime-$IMAGE" --build-arg OS_VERSION=16.04 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=${PYTHON_VER} -f Dockerfile.ubuntu .
-  fi
-else
-  IMAGE="centos7"
-  PYTHON_VER=3.6
-  echo "Image used for testing is onnxruntime-$IMAGE"
 
-  sudo docker build -t "onnxruntime-$IMAGE" --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=${PYTHON_VER} -f Dockerfile.centos .
-fi
-
-sudo docker rm -f "onnxruntime-cpu" || true
-
-
-sudo --preserve-env docker run -h $HOSTNAME \
-        --rm \
+docker run --rm \
         --name "onnxruntime-cpu" \
         --volume "$SOURCE_ROOT:/onnxruntime_src" \
         --volume "$BUILD_DIR:/home/onnxruntimedev" \
@@ -53,6 +35,6 @@ sudo --preserve-env docker run -h $HOSTNAME \
         -e "DisableMlOps=$DISABLEMLOPS" \
         -e "RunTestCsharp=$RunTestCsharp" \
         -e "RunTestNative=$RunTestNative" \
-        "onnxruntime-$IMAGE" \
+        onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn \
         /bin/bash /onnxruntime_src/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest.sh \
         /home/onnxruntimedev/$NUGET_REPO_DIRNAME /onnxruntime_src /home/onnxruntimedev $CurrentOnnxRuntimeVersion
diff --git a/onnxruntime/core/optimizer/matmul_scale_fusion.cc b/onnxruntime/core/optimizer/matmul_scale_fusion.cc
index 7356f55177..2b540d691b 100644
--- a/onnxruntime/core/optimizer/matmul_scale_fusion.cc
+++ b/onnxruntime/core/optimizer/matmul_scale_fusion.cc
@@ -43,7 +43,7 @@ optional<float> GetScalarConstantInitializer(const Graph& graph, const NodeArg&
     return {};
   }
 
-  float scalar;
+  float scalar=0.f;
   utils::MLTypeCallDispatcherRet<
       Status, ExtractScalarAsFloatDispatchTarget,
       uint32_t, uint64_t, int32_t, int64_t, MLFloat16, float, double>
diff --git a/onnxruntime/test/tvm/tvm_basic_test.cc b/onnxruntime/test/tvm/tvm_basic_test.cc
index 59784caadf..03b7a67d98 100644
--- a/onnxruntime/test/tvm/tvm_basic_test.cc
+++ b/onnxruntime/test/tvm/tvm_basic_test.cc
@@ -251,7 +251,7 @@ class FuseExecutionProviderX : public CPUExecutionProvider {
         tvm::TVMRetValue rvalue;
         try {
           evaluate_func_.CallPacked(tvm_args, &rvalue);
-        } catch (std::exception ex) {
+        } catch (std::exception& ex) {
           return Status(common::ONNXRUNTIME, common::FAIL);  // TODO: Translate exception to error code
         }
         if (rvalue.type_code() != kNull) {
diff --git a/setup.py b/setup.py
index 5fe14996da..6ef3cff5c8 100644
--- a/setup.py
+++ b/setup.py
@@ -157,7 +157,7 @@ except ImportError as error:
 
 # Additional binaries
 if platform.system() == 'Linux':
-  libs = ['onnxruntime_pybind11_state.so', 'libdnnl.so.1', 'libmklml_intel.so', 'libiomp5.so', 'mimalloc.so']
+  libs = ['onnxruntime_pybind11_state.so', 'libdnnl.so.1', 'libmklml_intel.so', 'libmklml_gnu.so', 'libiomp5.so', 'mimalloc.so']
   # DNNL & TensorRT EPs are built as shared libs
   libs.extend(['libonnxruntime_providers_shared.so'])
   libs.extend(['libonnxruntime_providers_dnnl.so'])
diff --git a/tools/ci_build/build.py b/tools/ci_build/build.py
index 207089c736..0eb506defa 100755
--- a/tools/ci_build/build.py
+++ b/tools/ci_build/build.py
@@ -265,12 +265,8 @@ def parse_arguments():
         "--use_preinstalled_eigen", action='store_true',
         help="Use pre-installed Eigen.")
     parser.add_argument("--eigen_path", help="Path to pre-installed Eigen.")
-    parser.add_argument(
-        "--use_tvm", action="store_true", help="Build with TVM")
     parser.add_argument(
         "--use_openmp", action='store_true', help="Build with OpenMP")
-    parser.add_argument(
-        "--use_llvm", action="store_true", help="Build TVM with LLVM")
     parser.add_argument(
         "--enable_msinternal", action="store_true",
         help="Enable for Microsoft internal builds only.")
@@ -584,31 +580,16 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
         "-Donnxruntime_USE_DNNL=" + ("ON" if args.use_dnnl else "OFF"),
         "-Donnxruntime_USE_MKLML=" + ("ON" if args.use_mklml else "OFF"),
         "-Donnxruntime_USE_NGRAPH=" + ("ON" if args.use_ngraph else "OFF"),
-        "-Donnxruntime_USE_OPENVINO=" + ("ON" if args.use_openvino else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_MYRIAD=" + (
-            "ON" if args.use_openvino == "MYRIAD_FP16" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_GPU_FP32=" + (
-            "ON" if args.use_openvino == "GPU_FP32" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_GPU_FP16=" + (
-            "ON" if args.use_openvino == "GPU_FP16" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_CPU_FP32=" + (
-            "ON" if args.use_openvino == "CPU_FP32" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_VAD_M=" + (
-            "ON" if args.use_openvino == "VAD-M_FP16" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_VAD_F=" + (
-            "ON" if args.use_openvino == "VAD-F_FP32" else "OFF"),
-        "-Donnxruntime_USE_OPENVINO_BINARY=" + (
-            "ON" if args.use_openvino else "OFF"),
         "-Donnxruntime_USE_NNAPI_BUILTIN=" + ("ON" if args.use_nnapi else "OFF"),
         "-Donnxruntime_USE_RKNPU=" + ("ON" if args.use_rknpu else "OFF"),
         "-Donnxruntime_USE_OPENMP=" + (
             "ON" if args.use_openmp and not (
-                args.use_nnapi or args.use_mklml or args.use_ngraph or
+                args.use_nnapi or (args.use_mklml and (is_macOS() or is_windows())) or args.use_ngraph or
                 args.android or (args.ios and is_macOS())
                 or args.use_rknpu)
             else "OFF"),
-        "-Donnxruntime_USE_TVM=" + ("ON" if args.use_tvm else "OFF"),
-        "-Donnxruntime_USE_LLVM=" + ("ON" if args.use_llvm else "OFF"),
+        "-Donnxruntime_USE_TVM=" + ("ON" if args.use_nuphar else "OFF"),
+        "-Donnxruntime_USE_LLVM=" + ("ON" if args.use_nuphar else "OFF"),
         "-Donnxruntime_ENABLE_MICROSOFT_INTERNAL=" + (
             "ON" if args.enable_msinternal else "OFF"),
         "-Donnxruntime_USE_VITISAI=" + ("ON" if args.use_vitisai else "OFF"),
@@ -674,7 +655,22 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
     if args.winml_root_namespace_override:
         cmake_args += ["-Donnxruntime_WINML_NAMESPACE_OVERRIDE=" +
                        args.winml_root_namespace_override]
-
+    if args.use_openvino:
+        cmake_args += ["-Donnxruntime_USE_OPENVINO=ON",
+                       "-Donnxruntime_USE_OPENVINO_MYRIAD=" + (
+                           "ON" if args.use_openvino == "MYRIAD_FP16" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_GPU_FP32=" + (
+                           "ON" if args.use_openvino == "GPU_FP32" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_GPU_FP16=" + (
+                           "ON" if args.use_openvino == "GPU_FP16" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_CPU_FP32=" + (
+                           "ON" if args.use_openvino == "CPU_FP32" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_VAD_M=" + (
+                           "ON" if args.use_openvino == "VAD-M_FP16" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_VAD_F=" + (
+                           "ON" if args.use_openvino == "VAD-F_FP32" else "OFF"),
+                       "-Donnxruntime_USE_OPENVINO_BINARY=" + (
+                           "ON" if args.use_openvino else "OFF")]
     # temp turn on only for linux gpu build
     if not is_windows():
         if args.use_cuda:
@@ -690,7 +686,7 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
             "-DProtobuf_USE_STATIC_LIBS=ON"
         ]
 
-    if args.use_llvm:
+    if args.use_nuphar and args.llvm_path is not None:
         cmake_args += ["-DLLVM_DIR=%s" % args.llvm_path]
 
     if args.use_cuda and not is_windows():
@@ -709,12 +705,6 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
             "-DANDROID_ABI=" + str(args.android_abi)
         ]
 
-    if is_macOS():
-        if args.use_xcode:
-            cmake_args += ['-G', 'Xcode']
-        elif args.cmake_generator is not None:
-            cmake_args += ['-G', args.cmake_generator]
-
     if args.ios:
         if is_macOS():
             needed_args = [
@@ -813,8 +803,7 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
 
     cmake_args += ["-D{}".format(define) for define in cmake_extra_defines]
 
-    if is_windows():
-        cmake_args += cmake_extra_args
+    cmake_args += cmake_extra_args
 
     # ADO pipelines will store the pipeline build number
     # (e.g. 191101-2300.1.master) and source version in environment
@@ -858,16 +847,15 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
     for config in configs:
         config_build_dir = get_config_build_dir(build_dir, config)
         os.makedirs(config_build_dir, exist_ok=True)
-
-        if args.use_tvm:
+        if args.use_nuphar:
             os.environ["PATH"] = os.path.join(
                 config_build_dir, "external", "tvm",
-                config) + os.pathsep + os.environ["PATH"]
+                config) + os.pathsep + os.path.dirname(sys.executable) + os.pathsep + os.environ["PATH"]
 
         run_subprocess(
             cmake_args + [
                 "-Donnxruntime_ENABLE_MEMLEAK_CHECKER=" +
-                ("ON" if config.lower() == 'debug' and not args.use_tvm and not
+                ("ON" if config.lower() == 'debug' and not args.use_nuphar and not
                  args.use_ngraph and not args.use_openvino and not
                  args.enable_msvc_static_runtime
                  else "OFF"), "-DCMAKE_BUILD_TYPE={}".format(config)],
@@ -906,7 +894,7 @@ def build_targets(args, cmake_path, build_dir, configs, parallel):
             elif (is_macOS() and args.use_xcode):
                 # CMake will generate correct build tool args for Xcode
                 cmd_args += ["--parallel", num_cores]
-            else:
+            elif args.cmake_generator != 'Ninja':
                 build_tool_args += ["-j" + num_cores]
 
         if build_tool_args:
@@ -1174,13 +1162,16 @@ def run_onnxruntime_tests(args, source_dir, ctest_path, build_dir, configs):
                     'cd /data/local/tmp && /data/local/tmp/onnx_test_runner /data/local/tmp/test')  # noqa
             continue
         dll_path_list = []
-        if args.use_tvm:
+        if args.use_nuphar:
             dll_path_list.append(os.path.join(
                 build_dir, config, "external", "tvm", config))
         if args.use_tensorrt:
             dll_path_list.append(os.path.join(args.tensorrt_home, 'lib'))
         if args.use_mklml:
             dll_path_list.append(os.path.join(build_dir, config, "mklml", "src", "project_mklml", "lib"))
+        if not is_windows():
+            # A workaround for making libonnxruntime_providers_shared.so loadable.
+            dll_path_list.append(os.path.join(build_dir, config))
 
         dll_path = None
         if len(dll_path_list) > 0:
@@ -1407,16 +1398,14 @@ def build_protoc_for_host(cmake_path, source_dir, build_dir, args):
     ]
 
     is_ninja = args.cmake_generator == 'Ninja'
-
+    if args.cmake_generator is not None and not (is_macOS() and args.use_xcode):
+        cmd_args += ['-G', args.cmake_generator]
     if is_windows():
         if not is_ninja:
             cmd_args += ['-T', 'host=x64']
-        cmd_args += ['-G', args.cmake_generator]
     elif is_macOS():
         if args.use_xcode:
             cmd_args += ['-G', 'Xcode']
-        elif args.cmake_generator is not None:
-            cmd_args += ['-G', args.cmake_generator]
 
     run_subprocess(cmd_args, cwd=protoc_build_dir)
     # Build step
@@ -1614,6 +1603,10 @@ def main():
                 cmake_extra_args.append(
                     '-DCMAKE_TOOLCHAIN_FILE=' + os.path.join(
                         source_dir, 'cmake', 'wcos_toolchain.cmake'))
+        elif args.cmake_generator is not None and not (is_macOS() and args.use_xcode):
+            cmake_extra_args += ['-G', args.cmake_generator]
+        elif is_macOS() and args.use_xcode:
+            cmake_extra_args += ['-G', 'Xcode']
 
         if (args.android or args.ios) and args.path_to_protoc_exe is None:
             # Cross-compiling for Android and iOS
diff --git a/tools/ci_build/github/azure-pipelines/c-api-packaging-pipelines.yml b/tools/ci_build/github/azure-pipelines/c-api-packaging-pipelines.yml
index 529aa3756f..cfde7c25d8 100644
--- a/tools/ci_build/github/azure-pipelines/c-api-packaging-pipelines.yml
+++ b/tools/ci_build/github/azure-pipelines/c-api-packaging-pipelines.yml
@@ -7,18 +7,27 @@ jobs:
     vmImage: 'ubuntu-latest'
   steps:
     - template: templates/set-version-number-variables-step.yml
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos6 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos6 .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-    - task: CmdLine@2
-      inputs:
-        script: |
+          mkdir -p $HOME/.onnx
           docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro \
-          -e NIGHTLY_BUILD onnxruntime-centos6 /usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib \
-          --use_openmp --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest
+          --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /opt/python/cp37-cp37m/bin/python3 \
+          /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --enable_lto \
+          --skip_submodule_sync  --parallel --build_shared_lib --use_openmp
         workingDirectory: $(Build.SourcesDirectory)
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - template: templates/c-api-artifacts-package-and-publish-steps-posix.yml
       parameters:
           buildConfig: 'Release'
diff --git a/tools/ci_build/github/azure-pipelines/java-api-packaging-pipelines.yml b/tools/ci_build/github/azure-pipelines/java-api-packaging-pipelines.yml
index 9eaa40f772..291bd924f7 100644
--- a/tools/ci_build/github/azure-pipelines/java-api-packaging-pipelines.yml
+++ b/tools/ci_build/github/azure-pipelines/java-api-packaging-pipelines.yml
@@ -16,19 +16,25 @@ jobs:
     - template: templates/set-version-number-variables-step.yml
     - template: templates/linux-set-variables-and-download.yml
 
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos6 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos6 .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-        displayName: 'Build CentoOS6 docker image'
-    - task: CmdLine@2
-      inputs:
-        script: |
-          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro -e NIGHTLY_BUILD onnxruntime-centos6 /bin/bash -c "/usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib --build_java --use_openmp --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest --enable_onnx_tests && cd /build/Release && make install DESTDIR=/build/linux-x64"
+          mkdir -p $HOME/.onnx
+          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /bin/bash -c "/opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib --build_java --use_openmp --enable_onnx_tests && cd /build/Release && make install DESTDIR=/build/linux-x64"
         workingDirectory: $(Build.SourcesDirectory)
         displayName: 'Run build and test'
-
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - template: templates/java-api-artifacts-package-and-publish-steps-posix.yml
       parameters:
           arch: 'linux-x64'
diff --git a/tools/ci_build/github/azure-pipelines/linux-ci-pipeline.yml b/tools/ci_build/github/azure-pipelines/linux-ci-pipeline.yml
index a65b0ad991..a4d75e8f38 100644
--- a/tools/ci_build/github/azure-pipelines/linux-ci-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/linux-ci-pipeline.yml
@@ -1,9 +1,63 @@
 jobs:
-- template: templates/linux-ci.yml
-  parameters:
-    AgentPool : 'Linux-CPU'
-    JobName: 'Linux_CI_Dev'
-    BuildCommand: 'tools/ci_build/github/linux/run_dockerbuild.sh -o ubuntu16.04 -d cpu -r $(Build.BinariesDirectory) -x "--use_mklml --use_llvm --use_nuphar --use_dnnl --use_tvm --build_wheel --build_java --build_nodejs --enable_language_interop_ops"'
-    DoNugetPack:  'false'
-    ArtifactName: 'drop-linux'
-    TimeoutInMinutes: 180
+- job: Linux_Build
+  timeoutInMinutes: 120
+  workspace:
+    clean: all
+  pool: Linux-CPU
+  steps:
+  - checkout: self
+    clean: true
+    submodules: recursive
+
+  - task: Docker@2
+    displayName: login
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: login
+      addPipelineData: false
+  - task: CmdLine@2
+    inputs:
+      script: |
+        mkdir -p $HOME/.onnx
+        docker run --rm \
+          --volume /data/onnx:/data/onnx:ro \
+          --volume $(Build.SourcesDirectory):/onnxruntime_src \
+          --volume $(Build.BinariesDirectory):/build \
+          --volume /data/models:/build/models:ro \
+          --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
+          -e NIGHTLY_BUILD \
+          -e BUILD_BUILDNUMBER \
+          onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn \
+            /opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
+              --build_dir /build --cmake_generator Ninja \
+              --config Debug Release \
+              --skip_submodule_sync \
+              --build_shared_lib \
+              --parallel \
+              --build_wheel \
+              --use_openmp \
+              --enable_onnx_tests \
+              --use_mklml --enable_pybind --build_java --build_nodejs \
+              --cmake_extra_defines PYTHON_INCLUDE_DIR=/opt/python/cp37-cp37m/include/python3.7m PYTHON_LIBRARY=/usr/lib64/librt.so
+      workingDirectory: $(Build.SourcesDirectory)
+  - task: Docker@2
+    displayName: logout
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: logout
+      addPipelineData: false
+  - task: PublishTestResults@2
+    displayName: 'Publish unit test results'
+    inputs:
+      testResultsFiles: '**/*.results.xml'
+      searchFolder: '$(Build.BinariesDirectory)'
+      testRunTitle: 'Unit Test Run'
+    condition: succeededOrFailed()
+
+  - template: templates/component-governance-component-detection-steps.yml
+    parameters:
+      condition: 'succeeded'
+
+  - task: mspremier.PostBuildCleanup.PostBuildCleanup-task.PostBuildCleanup@3
+    displayName: 'Clean Agent Directories'
+    condition: always()
diff --git a/tools/ci_build/github/azure-pipelines/linux-dnnl-ci-pipeline.yml b/tools/ci_build/github/azure-pipelines/linux-dnnl-ci-pipeline.yml
new file mode 100644
index 0000000000..1d660d99e4
--- /dev/null
+++ b/tools/ci_build/github/azure-pipelines/linux-dnnl-ci-pipeline.yml
@@ -0,0 +1,63 @@
+jobs:
+- job: Linux_py_Wheels
+  timeoutInMinutes: 180
+  workspace:
+    clean: all
+  pool: Linux-CPU
+  steps:
+  - checkout: self
+    clean: true
+    submodules: recursive
+
+  - task: Docker@2
+    displayName: login
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: login
+      addPipelineData: false
+  - task: CmdLine@2
+    inputs:
+      script: |
+        mkdir -p $HOME/.onnx
+        docker run --rm \
+          --volume /data/onnx:/data/onnx:ro \
+          --volume $(Build.SourcesDirectory):/onnxruntime_src \
+          --volume $(Build.BinariesDirectory):/build \
+          --volume /data/models:/build/models:ro \
+          --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
+          -e NIGHTLY_BUILD \
+          -e BUILD_BUILDNUMBER \
+          onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn \
+            /opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
+              --build_dir /build \
+              --config Debug Release \
+              --skip_submodule_sync \
+              --build_shared_lib \
+              --parallel \
+              --enable_pybind \
+              --use_openmp \
+              --enable_onnx_tests \
+              --use_dnnl \
+              --cmake_extra_defines PYTHON_INCLUDE_DIR=/opt/python/cp37-cp37m/include/python3.7m PYTHON_LIBRARY=/usr/lib64/librt.so
+      workingDirectory: $(Build.SourcesDirectory)
+  - task: Docker@2
+    displayName: logout
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: logout
+      addPipelineData: false
+  - task: PublishTestResults@2
+    displayName: 'Publish unit test results'
+    inputs:
+      testResultsFiles: '**/*.results.xml'
+      searchFolder: '$(Build.BinariesDirectory)'
+      testRunTitle: 'Unit Test Run'
+    condition: succeededOrFailed()
+
+  - template: templates/component-governance-component-detection-steps.yml
+    parameters:
+      condition: 'succeeded'
+
+  - task: mspremier.PostBuildCleanup.PostBuildCleanup-task.PostBuildCleanup@3
+    displayName: 'Clean Agent Directories'
+    condition: always()
diff --git a/tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml b/tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml
index cbadf2c4e0..e2f8a4f08b 100644
--- a/tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml
@@ -1,8 +1,63 @@
 jobs:
-- template: templates/linux-ci.yml
-  parameters:
-    AgentPool : 'Linux-GPU-CUDA10'
-    JobName: 'Linux_CI_GPU_Dev'
-    BuildCommand: 'tools/ci_build/github/linux/run_dockerbuild.sh -o ubuntu16.04 -d gpu -r $(Build.BinariesDirectory) -x "--build_wheel"'
-    DoNugetPack:  'false'
-    ArtifactName: 'drop-linux'
+- job: Linux_Build
+  timeoutInMinutes: 120
+  workspace:
+    clean: all
+  pool: Linux-GPU-CUDA10
+  steps:
+  - checkout: self
+    clean: true
+    submodules: recursive
+
+  - task: Docker@2
+    displayName: login
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: login
+      addPipelineData: false
+  - task: CmdLine@2
+    inputs:
+      script: |
+        mkdir -p $HOME/.onnx
+        docker run --gpus all --rm \
+          --volume /data/onnx:/data/onnx:ro \
+          --volume $(Build.SourcesDirectory):/onnxruntime_src \
+          --volume $(Build.BinariesDirectory):/build \
+          --volume /data/models:/build/models:ro \
+          --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
+          -e NIGHTLY_BUILD \
+          -e BUILD_BUILDNUMBER \
+          onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chp \
+            /opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
+              --build_dir /build --cmake_generator Ninja \
+              --config Debug Release \
+              --skip_submodule_sync \
+              --build_shared_lib \
+              --parallel \
+              --build_wheel \
+              --use_openmp \
+              --enable_onnx_tests --use_cuda --cuda_version=10.1 --cuda_home=/usr/local/cuda-10.1 --cudnn_home=/usr/local/cuda-10.1 \
+              --use_mklml --enable_pybind --build_java --build_nodejs \
+              --cmake_extra_defines PYTHON_INCLUDE_DIR=/opt/python/cp37-cp37m/include/python3.7m PYTHON_LIBRARY=/usr/lib64/librt.so
+      workingDirectory: $(Build.SourcesDirectory)
+  - task: Docker@2
+    displayName: logout
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: logout
+      addPipelineData: false
+  - task: PublishTestResults@2
+    displayName: 'Publish unit test results'
+    inputs:
+      testResultsFiles: '**/*.results.xml'
+      searchFolder: '$(Build.BinariesDirectory)'
+      testRunTitle: 'Unit Test Run'
+    condition: succeededOrFailed()
+
+  - template: templates/component-governance-component-detection-steps.yml
+    parameters:
+      condition: 'succeeded'
+
+  - task: mspremier.PostBuildCleanup.PostBuildCleanup-task.PostBuildCleanup@3
+    displayName: 'Clean Agent Directories'
+    condition: always()
diff --git a/tools/ci_build/github/azure-pipelines/linux-nuphar-ci-pipeline.yml b/tools/ci_build/github/azure-pipelines/linux-nuphar-ci-pipeline.yml
new file mode 100644
index 0000000000..5fb4244f3c
--- /dev/null
+++ b/tools/ci_build/github/azure-pipelines/linux-nuphar-ci-pipeline.yml
@@ -0,0 +1,9 @@
+jobs:
+- template: templates/linux-ci.yml
+  parameters:
+    AgentPool : 'Linux-CPU'
+    JobName: 'Linux_CI_Dev'
+    BuildCommand: 'tools/ci_build/github/linux/run_dockerbuild.sh -o ubuntu16.04 -d cpu -r $(Build.BinariesDirectory) -x "--enable_pybind --use_nuphar"'
+    DoNugetPack:  'false'
+    ArtifactName: 'drop-linux'
+    TimeoutInMinutes: 180
\ No newline at end of file
diff --git a/tools/ci_build/github/azure-pipelines/nodejs/templates/cpu.yml b/tools/ci_build/github/azure-pipelines/nodejs/templates/cpu.yml
index 729fdae8b5..6edf95bfbe 100644
--- a/tools/ci_build/github/azure-pipelines/nodejs/templates/cpu.yml
+++ b/tools/ci_build/github/azure-pipelines/nodejs/templates/cpu.yml
@@ -38,16 +38,26 @@ jobs:
   steps:    
     - template: ../../templates/set-version-number-variables-step.yml
     - template: ../../templates/linux-set-variables-and-download.yml
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos7 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-    - task: CmdLine@2
-      inputs:
-        script: |
-          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro -e NIGHTLY_BUILD onnxruntime-centos7 /bin/bash -c "/usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_nodejs --use_openmp --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest --enable_onnx_tests && cd /onnxruntime_src/nodejs && npm pack"
+          mkdir -p $HOME/.onnx && docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro \
+          --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /bin/bash -c "/opt/python/cp37-cp37m/bin/python3 \
+          /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --enable_lto --build_nodejs \
+          --skip_submodule_sync  --parallel --build_shared_lib --use_openmp && cd /onnxruntime_src/nodejs && npm pack"
         workingDirectory: $(Build.SourcesDirectory)
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - script: |
        set -e -x
        cp $(Build.SourcesDirectory)/nodejs/onnxruntime-*.tgz $(Build.ArtifactStagingDirectory)
diff --git a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-mklml.yml b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-mklml.yml
index 635ac12a9b..5b33ee7481 100644
--- a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-mklml.yml
+++ b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-mklml.yml
@@ -30,21 +30,28 @@ jobs:
   steps:
     - template: ../../templates/set-version-number-variables-step.yml
     - template: ../../templates/linux-set-variables-and-download.yml
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos6 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos6 .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-    - task: CmdLine@2
-      inputs:
-        script: |
-          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro -e NIGHTLY_BUILD onnxruntime-centos6 /bin/bash -c "/usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest --use_mklml --enable_onnx_tests && cd /build/Release && make install DESTDIR=/build/linux-x64"
+          mkdir -p $HOME/.onnx
+          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /bin/bash -c "/opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --enable_lto --skip_submodule_sync --parallel --build_shared_lib --use_openmp --enable_onnx_tests --use_mklml && cd /build/Release && make install DESTDIR=/build/linux-x64"
         workingDirectory: $(Build.SourcesDirectory)
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - script: |
        set -e -x
        mv $(Build.BinariesDirectory)/linux-x64/usr/local/lib64 $(Build.BinariesDirectory)/linux-x64/linux-x64
-       cp $(Build.BinariesDirectory)/Release/mklml/src/project_mklml/lib/libiomp5.so $(Build.BinariesDirectory)/linux-x64/linux-x64
-       cp $(Build.BinariesDirectory)/Release/mklml/src/project_mklml/lib/libmklml_intel.so $(Build.BinariesDirectory)/linux-x64/linux-x64
+       cp $(Build.BinariesDirectory)/Release/mklml/src/project_mklml/lib/libmklml_gnu.so $(Build.BinariesDirectory)/linux-x64/linux-x64
        ldd $(Build.BinariesDirectory)/linux-x64/linux-x64/libonnxruntime.so
        cd $(Build.BinariesDirectory)/linux-x64
        zip -r linux-x64.zip linux-x64
@@ -140,8 +147,7 @@ jobs:
              mkdir %%~ni\runtimes\linux-x64
              mkdir %%~ni\runtimes\linux-x64\native
              move linux-x64\linux-x64\libonnxruntime.so %%~ni\runtimes\linux-x64\native\libonnxruntime.so
-             move linux-x64\linux-x64\libiomp5.so %%~ni\runtimes\linux-x64\native\libiomp5.so
-             move linux-x64\linux-x64\libmklml_intel.so %%~ni\runtimes\linux-x64\native\libmklml_intel.so
+             move linux-x64\linux-x64\libmklml_gnu.so %%~ni\runtimes\linux-x64\native\libmklml_gnu.so
              unzip osx-x64.zip -d osx-x64
              dir osx-x64 /s
              mkdir %%~ni\runtimes\osx-x64
diff --git a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-nocontribops-arm64.yml b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-nocontribops-arm64.yml
index f31670be80..bed1860fc8 100644
--- a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-nocontribops-arm64.yml
+++ b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu-nocontribops-arm64.yml
@@ -72,16 +72,24 @@ jobs:
   steps:    
     - template: ../../templates/set-version-number-variables-step.yml
     - template: ../../templates/linux-set-variables-and-download.yml
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos6 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos6 .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-    - task: CmdLine@2
-      inputs:
-        script: |
-          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro -e NIGHTLY_BUILD onnxruntime-centos6 /bin/bash -c "/usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest --enable_onnx_tests --disable_contrib_ops --disable_ml_ops && cd /build/Release && make install DESTDIR=/build/linux-x64"
+          mkdir -p $HOME/.onnx
+          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /bin/bash -c "/opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --enable_lto --skip_submodule_sync --parallel --build_shared_lib --enable_onnx_tests  --disable_contrib_ops --disable_ml_ops && cd /build/Release && make install DESTDIR=/build/linux-x64"
         workingDirectory: $(Build.SourcesDirectory)
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - script: |
        set -e -x
        mv $(Build.BinariesDirectory)/linux-x64/usr/local/lib64 $(Build.BinariesDirectory)/linux-x64/linux-x64
diff --git a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu.yml b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu.yml
index 42cc8e2ca4..c09acba23b 100644
--- a/tools/ci_build/github/azure-pipelines/nuget/templates/cpu.yml
+++ b/tools/ci_build/github/azure-pipelines/nuget/templates/cpu.yml
@@ -95,16 +95,24 @@ jobs:
   steps:    
     - template: ../../templates/set-version-number-variables-step.yml
     - template: ../../templates/linux-set-variables-and-download.yml
+    - task: Docker@2
+      displayName: login
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: login
+        addPipelineData: false
     - task: CmdLine@2
       inputs:
         script: |
-          docker build --pull -t onnxruntime-centos6 --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=3.6 -f Dockerfile.centos6 .
-        workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-    - task: CmdLine@2
-      inputs:
-        script: |
-          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro -e NIGHTLY_BUILD onnxruntime-centos6 /bin/bash -c "/usr/bin/python3.6 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --skip_submodule_sync  --parallel --build_shared_lib --use_openmp --cmake_path /usr/bin/cmake --ctest_path /usr/bin/ctest --enable_onnx_tests && cd /build/Release && make install DESTDIR=/build/linux-x64"
+          mkdir -p $HOME/.onnx
+          docker run --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build --volume /data/models:/build/models:ro --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn /bin/bash -c "/opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release --enable_lto --skip_submodule_sync --parallel --build_shared_lib --use_openmp --enable_onnx_tests && cd /build/Release && make install DESTDIR=/build/linux-x64"
         workingDirectory: $(Build.SourcesDirectory)
+    - task: Docker@2
+      displayName: logout
+      inputs:
+        containerRegistry: onnxruntimeregistry
+        command: logout
+        addPipelineData: false
     - script: |
        set -e -x
        mv $(Build.BinariesDirectory)/linux-x64/usr/local/lib64 $(Build.BinariesDirectory)/linux-x64/linux-x64
diff --git a/tools/ci_build/github/azure-pipelines/nuget/templates/test_linux.yml b/tools/ci_build/github/azure-pipelines/nuget/templates/test_linux.yml
index d37efd2cb2..2c10b8e8c8 100644
--- a/tools/ci_build/github/azure-pipelines/nuget/templates/test_linux.yml
+++ b/tools/ci_build/github/azure-pipelines/nuget/templates/test_linux.yml
@@ -34,6 +34,12 @@ jobs:
   - template: get-nuget-package-version-as-variable.yml
     parameters:
       packageFolder: '$(Build.BinariesDirectory)/nuget-artifact'
+  - task: Docker@2
+    displayName: login
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: login
+      addPipelineData: false
 
   - ${{ if eq(parameters['TestGPU'], 'false') }}:
     - script: |
@@ -43,13 +49,6 @@ jobs:
       env:
         OnnxRuntimeBuildDirectory: $(Build.BinariesDirectory)
 
-  - ${{ if eq(parameters['TestGPU'], 'false') }}:
-    - script: |
-       set -e -x
-       $(Build.SourcesDirectory)/csharp/test/Microsoft.ML.OnnxRuntime.EndToEndTests/runtest-docker.sh $(Build.SourcesDirectory) $(Build.BinariesDirectory) nuget-artifact $(NuGetPackageVersionNumber)
-      displayName: 'Run Package Test (x64) on Ubuntu'
-      env:
-        OnnxRuntimeBuildDirectory: $(Build.BinariesDirectory)
 
   - ${{ if eq(parameters['TestGPU'], 'true') }}:
     - script: |
@@ -59,6 +58,13 @@ jobs:
       env:
         OnnxRuntimeBuildDirectory: $(Build.BinariesDirectory)
 
+  - task: Docker@2
+    displayName: logout
+    inputs:
+      containerRegistry: onnxruntimeregistry
+      command: logout
+      addPipelineData: false
+
   - template: ../../templates/component-governance-component-detection-steps.yml
     parameters :
       condition : 'always'
diff --git a/tools/ci_build/github/azure-pipelines/orttraining-py-packaging-pipeline.yml b/tools/ci_build/github/azure-pipelines/orttraining-py-packaging-pipeline.yml
index b3b5e84f49..c435a7f698 100644
--- a/tools/ci_build/github/azure-pipelines/orttraining-py-packaging-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/orttraining-py-packaging-pipeline.yml
@@ -5,7 +5,8 @@ stages:
   parameters:
     build_py_parameters: --enable_training --wheel_name_suffix=training
     enable_linux_cpu: false
-    enable_linux_gpu: true
+    enable_linux_gpu: false
+    enable_linux_gpu_training: true
     # Python 3.5 build has issues with nightly torch and torchvision dependencies
     enable_linux_gpu_py35: false
     enable_windows_cpu: false
diff --git a/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml b/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml
index 7bc444daa7..f5334ea751 100644
--- a/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml
+++ b/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml
@@ -15,6 +15,11 @@ parameters:
   type: boolean
   default: true
 
+- name: enable_linux_gpu_training
+  displayName: 'Whether Linux GPU package is built.'
+  type: boolean
+  default: false
+
 - name: enable_linux_gpu_py35
   displayName: >
     Whether Linux GPU package is built for Python 3.5.
@@ -81,39 +86,43 @@ stages:
 
       - template: set-py-packaging-variables-step.yml
 
+      - task: Docker@2
+        displayName: login
+        inputs:
+          containerRegistry: onnxruntimeregistry
+          command: login
+          addPipelineData: false
       - task: CmdLine@2
         inputs:
           script: |
-            docker build --pull \
-              -t ${{ parameters.docker_image_prefix }}-manylinux-$(python.version) \
-              --build-arg BUILD_USER=onnxruntimedev \
-              --build-arg BUILD_UID=$(id -u) \
-              --build-arg PYTHON_VERSION=$(python.version) \
-              -f Dockerfile.manylinux2010_cpu .
-          workingDirectory: $(Build.SourcesDirectory)/tools/ci_build/github/linux/docker
-      - task: CmdLine@2
-        inputs:
-          script: |
+            mkdir -p $HOME/.onnx
             docker run --rm \
               --volume /data/onnx:/data/onnx:ro \
               --volume $(Build.SourcesDirectory):/onnxruntime_src \
               --volume $(Build.BinariesDirectory):/build \
               --volume /data/models:/build/models:ro \
+              --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
               -e NIGHTLY_BUILD \
               -e BUILD_BUILDNUMBER \
-              ${{ parameters.docker_image_prefix }}-manylinux-$(python.version) \
+              onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chn \
                 $(python.manylinux.dir)/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
-                  --build_dir /build \
+                  --build_dir /build --cmake_generator Ninja \
                   --config Release \
                   --skip_submodule_sync \
                   --parallel \
+                  --enable_lto \
                   --build_wheel \
                   --use_openmp \
                   --enable_onnx_tests \
                   ${{ parameters.build_py_parameters }} \
                   --cmake_extra_defines PYTHON_INCLUDE_DIR=$(python.manylinux.include.dir) PYTHON_LIBRARY=/usr/lib64/librt.so
           workingDirectory: $(Build.SourcesDirectory)
-
+      - task: Docker@2
+        displayName: logout
+        inputs:
+          containerRegistry: onnxruntimeregistry
+          command: logout
+          addPipelineData: false
       - task: CopyFiles@2
         displayName: 'Copy Python Wheel to: $(Build.ArtifactStagingDirectory)'
         inputs:
@@ -140,9 +149,85 @@ stages:
       pool: Linux-GPU-CUDA10
       strategy:
         matrix:
-          ${{ if eq(parameters.enable_linux_gpu_py35, true) }}:
-            Python35:
-              python.version: '3.5'
+          Python35:
+            python.version: '3.5'
+          Python36:
+            python.version: '3.6'
+          Python37:
+            python.version: '3.7'
+          Python38:
+            python.version: '3.8'
+      steps:
+      - checkout: self
+        clean: true
+        submodules: recursive
+
+      - template: set-py-packaging-variables-step.yml
+
+      - task: Docker@2
+        displayName: login
+        inputs:
+          containerRegistry: onnxruntimeregistry
+          command: login
+          addPipelineData: false
+      - task: CmdLine@2
+        inputs:
+          script: |
+            mkdir -p $HOME/.onnx
+            docker run --gpus all --rm \
+              --volume /data/onnx:/data/onnx:ro \
+              --volume $(Build.SourcesDirectory):/onnxruntime_src \
+              --volume $(Build.BinariesDirectory):/build \
+              --volume /data/models:/build/models:ro \
+              --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
+              -e NIGHTLY_BUILD \
+              -e BUILD_BUILDNUMBER \
+              onnxruntimeregistry.azurecr.io/internal/azureml/onnxruntimecpubuild:chp \
+                $(python.manylinux.dir)/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
+                  --build_dir /build --cmake_generator Ninja \
+                  --config Release \
+                  --skip_submodule_sync \
+                  --parallel \
+                  --build_wheel \
+                  --use_openmp \
+                  --enable_onnx_tests --use_cuda --cuda_version=10.1 --cuda_home=/usr/local/cuda-10.1 --cudnn_home=/usr/local/cuda-10.1 \
+                  ${{ parameters.build_py_parameters }} \
+                  --cmake_extra_defines PYTHON_INCLUDE_DIR=$(python.manylinux.include.dir) PYTHON_LIBRARY=/usr/lib64/librt.so
+          workingDirectory: $(Build.SourcesDirectory)
+
+      - task: Docker@2
+        displayName: logout
+        inputs:
+          containerRegistry: onnxruntimeregistry
+          command: logout
+          addPipelineData: false
+
+      - task: CopyFiles@2
+        displayName: 'Copy Python Wheel to: $(Build.ArtifactStagingDirectory)'
+        inputs:
+          SourceFolder: '$(Build.BinariesDirectory)'
+          Contents: 'Release/dist/*.whl'
+          TargetFolder: '$(Build.ArtifactStagingDirectory)'
+
+      - task: PublishBuildArtifacts@1
+        displayName: 'Publish Artifact: ONNXRuntime python wheel'
+        inputs:
+          ArtifactName: onnxruntime_gpu
+
+      - template: component-governance-component-detection-steps.yml
+        parameters:
+          condition: 'succeeded'
+
+      - template: clean-agent-build-directory-step.yml
+
+  - ${{ if eq(parameters.enable_linux_gpu_training, true) }}:
+    - job: Linux_py_GPU_Wheels
+      timeoutInMinutes: 120
+      workspace:
+        clean: all
+      pool: Linux-GPU-CUDA10
+      strategy:
+        matrix:
           Python36:
             python.version: '3.6'
           Python37:
@@ -170,11 +255,13 @@ stages:
       - task: CmdLine@2
         inputs:
           script: |
+            mkdir -p $HOME/.onnx
             docker run --rm --gpus all \
               --volume /data/onnx:/data/onnx:ro \
               --volume $(Build.SourcesDirectory):/onnxruntime_src \
               --volume $(Build.BinariesDirectory):/build \
               --volume /data/models:/build/models:ro \
+              --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
               -e NVIDIA_VISIBLE_DEVICES=all \
               -e NIGHTLY_BUILD \
               -e BUILD_BUILDNUMBER \
diff --git a/tools/ci_build/github/linux/docker/Dockerfile.centos6 b/tools/ci_build/github/linux/docker/Dockerfile.centos6
deleted file mode 100644
index a52cf31928..0000000000
--- a/tools/ci_build/github/linux/docker/Dockerfile.centos6
+++ /dev/null
@@ -1,14 +0,0 @@
-FROM mcr.microsoft.com/dotnet-buildtools/prereqs:centos-6-50f0d02-20190918213956
-
-ADD scripts /tmp/scripts
-RUN cd /tmp/scripts && /tmp/scripts/install_centos.sh
-ENV JAVA_HOME /usr/lib/jvm/jre-1.8.0-openjdk.x86_64
-ENV PATH /opt/rh/devtoolset-2/root/usr/bin:${JAVA_HOME}/bin:${PATH}
-RUN /tmp/scripts/install_deps.sh -p 3.6 && rm -rf /tmp/scripts
-ENV PATH ${PATH}:/usr/local/gradle/bin
-
-ARG BUILD_UID=1000
-ARG BUILD_USER=onnxruntimedev
-RUN adduser --uid $BUILD_UID $BUILD_USER 
-WORKDIR /home/$BUILD_USER
-USER $BUILD_USER
diff --git a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_cpu b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_cpu
index ab16f24892..24d330b900 100644
--- a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_cpu
+++ b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_cpu
@@ -1,11 +1,11 @@
-FROM quay.io/pypa/manylinux2010_x86_64:2020-04-04-f427f46
+FROM quay.io/pypa/manylinux2010_x86_64:latest
 
-ARG PYTHON_VERSION
-ADD scripts /tmp/scripts
-RUN cd /tmp/scripts && /tmp/scripts/install_centos.sh && /tmp/scripts/install_deps.sh -p $PYTHON_VERSION && rm -rf /tmp/scripts
+ADD manylinux /tmp/scripts
+RUN cd /tmp/scripts && /tmp/scripts/install_centos.sh && /tmp/scripts/install_deps.sh  && rm -rf /tmp/scripts
 
 ARG BUILD_UID=1000
 ARG BUILD_USER=onnxruntimedev
-RUN adduser --uid $BUILD_UID $BUILD_USER 
+RUN adduser --uid $BUILD_UID $BUILD_USER
 WORKDIR /home/$BUILD_USER
 USER $BUILD_USER
+ENV PATH /usr/local/gradle/bin:/usr/local/dotnet:$PATH
diff --git a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_gpu b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_gpu
index c2fbb36d3a..fd4ac04d2a 100644
--- a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_gpu
+++ b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2010_gpu
@@ -1,14 +1,11 @@
-FROM quay.io/pypa/manylinux2010_x86_64:2020-04-04-f427f46
+FROM quay.io/pypa/manylinux2010_x86_64:latest
 
-ARG PYTHON_VERSION=3.5
+ADD manylinux /tmp/scripts
+RUN cd /tmp/scripts && /tmp/scripts/install_centos.sh && /tmp/scripts/install_deps.sh  && rm -rf /tmp/scripts
 
 ENV NVIDIA_VISIBLE_DEVICES all
 ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
 
-ADD scripts /tmp/scripts
-RUN cd /tmp/scripts && /tmp/scripts/install_centos.sh && /tmp/scripts/install_deps.sh -p $PYTHON_VERSION && \
-rm -rf /tmp/scripts
-
 #Below are copied from https://gitlab.com/nvidia/container-images/cuda/tree/master/dist/centos6
 
 RUN NVIDIA_GPGKEY_SUM=d1be581509378368edeec8c1eb2958702feedf3bc3d17011adbf24efacce4ab5 && \
@@ -67,3 +64,4 @@ ARG BUILD_USER=onnxruntimedev
 RUN adduser --comment 'onnxruntime Build User' --uid $BUILD_UID $BUILD_USER
 WORKDIR /home/$BUILD_USER
 USER $BUILD_USER
+ENV PATH /usr/local/gradle/bin:/usr/local/dotnet:$PATH
\ No newline at end of file
diff --git a/tools/ci_build/github/linux/docker/manylinux/install_centos.sh b/tools/ci_build/github/linux/docker/manylinux/install_centos.sh
new file mode 100755
index 0000000000..f4319a0c53
--- /dev/null
+++ b/tools/ci_build/github/linux/docker/manylinux/install_centos.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+set -e
+
+os_major_version=$(cat /etc/redhat-release | tr -dc '0-9.'|cut -d \. -f1)
+
+
+if ! rpm -q --quiet epel-release ; then
+  yum install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-$os_major_version.noarch.rpm
+fi
+
+echo "installing for os major version : $os_major_version"
+yum install -y redhat-lsb-core expat-devel libcurl-devel tar unzip curl zlib-devel make libunwind icu aria2 rsync bzip2 git bzip2-devel
+
+if [ "$os_major_version" == "6" ]; then
+    # See https://github.com/dotnet/core/blob/master/Documentation/build-and-install-rhel6-prerequisites.md
+    yum install -y epel-release libunwind openssl libnghttp2 libidn krb5-libs libuuid lttng-ust zlib
+    curl -sSL -o /tmp/1.tgz https://github.com/unicode-org/icu/releases/download/release-57-1/icu4c-57_1-RHEL6-x64.tgz
+    mkdir /tmp/icu
+    tar -zxf /tmp/1.tgz --strip=2 -C /tmp/icu
+    mv /tmp/icu/lib /tmp/icu/lib64
+    /bin/cp -r /tmp/icu/* /usr/
+    ldconfig /usr/lib64
+    rm -rf /tmp/icu
+    rm /tmp/1.tgz
+    
+    curl -o /tmp/d.sh -sSL https://dot.net/v1/dotnet-install.sh
+    chmod +x /tmp/d.sh
+    /tmp/d.sh --install-dir /usr/local/dotnet -c 2.1
+    rm /tmp/d.sh
+fi
+if [ "$os_major_version" == "7" ]; then
+    # install dotnet core dependencies
+    yum install -y lttng-ust openssl-libs krb5-libs libicu libuuid
+    # install dotnet runtimes
+    yum install -y https://packages.microsoft.com/config/centos/7/packages-microsoft-prod.rpm
+    yum install -y dotnet-sdk-2.1
+fi
+
+yum install -y java-1.8.0-openjdk-devel
+
diff --git a/tools/ci_build/github/linux/docker/manylinux/install_deps.sh b/tools/ci_build/github/linux/docker/manylinux/install_deps.sh
new file mode 100755
index 0000000000..f6cefa9190
--- /dev/null
+++ b/tools/ci_build/github/linux/docker/manylinux/install_deps.sh
@@ -0,0 +1,91 @@
+#!/bin/bash
+set -e -x
+
+#Download a file from internet
+function GetFile {
+  local uri=$1
+  local path=$2
+  local force=${3:-false}
+  local download_retries=${4:-5}
+  local retry_wait_time_seconds=${5:-30}
+
+  if [[ -f $path ]]; then
+    if [[ $force = false ]]; then
+      echo "File '$path' already exists. Skipping download"
+      return 0
+    else
+      rm -rf $path
+    fi
+  fi
+
+  if [[ -f $uri ]]; then
+    echo "'$uri' is a file path, copying file to '$path'"
+    cp $uri $path
+    return $?
+  fi
+
+  echo "Downloading $uri"
+  # Use aria2c if available, otherwise use curl
+  if command -v aria2c > /dev/null; then
+    aria2c -q -d $(dirname $path) -o $(basename $path) "$uri"
+  else
+    curl "$uri" -sSL --retry $download_retries --retry-delay $retry_wait_time_seconds --create-dirs -o "$path" --fail
+  fi
+
+  return $?
+}
+
+PYTHON_EXES=("/opt/python/cp35-cp35m/bin/python3.5" "/opt/python/cp36-cp36m/bin/python3.6" "/opt/python/cp37-cp37m/bin/python3.7" "/opt/python/cp38-cp38/bin/python3.8")
+
+SYS_LONG_BIT=$(getconf LONG_BIT)
+mkdir -p /tmp/src
+GLIBC_VERSION=$(getconf GNU_LIBC_VERSION | cut -f 2 -d \.)
+
+DISTRIBUTOR=$(lsb_release -i -s)
+
+if [[ "$DISTRIBUTOR" = "CentOS" && $SYS_LONG_BIT = "64" ]]; then
+  LIBDIR="lib64"
+else
+  LIBDIR="lib"
+fi
+
+cd /tmp/src
+
+  echo "Installing azcopy"
+  mkdir -p /tmp/azcopy
+  GetFile https://aka.ms/downloadazcopy-v10-linux /tmp/azcopy/azcopy.tar.gz
+  tar --strip 1 -xf /tmp/azcopy/azcopy.tar.gz -C /tmp/azcopy
+  cp /tmp/azcopy/azcopy /usr/bin
+  echo "Installing cmake"
+  GetFile https://github.com/Kitware/CMake/releases/download/v3.18.1/cmake-3.18.1-Linux-x86_64.tar.gz /tmp/src/cmake-3.18.1-Linux-x86_64.tar.gz
+  tar -zxf /tmp/src/cmake-3.18.1-Linux-x86_64.tar.gz --strip=1 -C /usr
+  echo "Installing Ninja"
+  GetFile https://github.com/ninja-build/ninja/archive/v1.10.0.tar.gz /tmp/src/ninja-linux.tar.gz
+  tar -zxf ninja-linux.tar.gz
+  cd ninja-1.10.0
+  cmake -Bbuild-cmake -H.
+  cmake --build build-cmake
+  mv ./build-cmake/ninja /usr/bin
+  echo "Installing Node.js"
+  GetFile https://nodejs.org/dist/v12.16.3/node-v12.16.3.tar.xz /tmp/src/node-v12.16.3.tar.xz
+  tar -xf /tmp/src/node-v12.16.3.tar.xz
+  cd node-v12.16.3
+  LDFLAGS=-lrt /opt/python/cp27-cp27m/bin/python configure --ninja
+  LDFLAGS=-lrt make -j$(getconf _NPROCESSORS_ONLN)
+  LDFLAGS=-lrt make install
+
+cd /tmp/src
+GetFile https://downloads.gradle-dn.com/distributions/gradle-6.3-bin.zip /tmp/src/gradle-6.3-bin.zip
+unzip /tmp/src/gradle-6.3-bin.zip
+mv /tmp/src/gradle-6.3 /usr/local/gradle
+
+
+for PYTHON_EXE in "${PYTHON_EXES[@]}"
+do
+  ${PYTHON_EXE} -m pip install -r ${0/%install_deps\.sh/requirements\.txt}
+done
+
+
+cd /
+rm -rf /tmp/src
+
diff --git a/tools/ci_build/github/linux/docker/manylinux/requirements.txt b/tools/ci_build/github/linux/docker/manylinux/requirements.txt
new file mode 100644
index 0000000000..2be05ad9f1
--- /dev/null
+++ b/tools/ci_build/github/linux/docker/manylinux/requirements.txt
@@ -0,0 +1,6 @@
+numpy==1.16.6
+mypy
+pytest
+setuptools>=41.4.0
+wheel
+onnx==1.7.0
\ No newline at end of file
diff --git a/tools/ci_build/github/linux/docker/scripts/install_server_deps.sh b/tools/ci_build/github/linux/docker/scripts/install_server_deps.sh
index ae88d26598..01d81ed68b 100755
--- a/tools/ci_build/github/linux/docker/scripts/install_server_deps.sh
+++ b/tools/ci_build/github/linux/docker/scripts/install_server_deps.sh
@@ -3,7 +3,6 @@ set -e
 
 SYS_LONG_BIT=$(getconf LONG_BIT)
 
-apt-get remove -y libprotobuf-dev libprotobuf9v5 protobuf-compiler libprotobuf-lite9v5
 apt-get update && apt-get install -y --no-install-recommends libre2-dev
 rm -rf /var/lib/apt/lists/*
 
@@ -15,12 +14,12 @@ if [ $SYS_LONG_BIT = "64" ]; then
   tar -C /usr/local -vzxf /tmp/go/go1.12.6.linux-amd64.tar.gz
 
   echo "Installing CMAKE"
-  aria2c https://github.com/Kitware/CMake/releases/download/v3.16.2/cmake-3.16.2-Linux-x86_64.tar.gz
-  tar -zxf cmake-3.16.2-Linux-x86_64.tar.gz --strip=1 -C /usr
+  aria2c https://github.com/Kitware/CMake/releases/download/v3.18.1/cmake-3.18.1-Linux-x86_64.tar.gz
+  tar -zxf cmake-3.18.1-Linux-x86_64.tar.gz --strip=1 -C /usr
 
   echo "Installing onnxruntime"
-  aria2c https://github.com/microsoft/onnxruntime/releases/download/v1.2.0/onnxruntime-linux-x64-1.2.0.tgz
-  tar -zxf onnxruntime-linux-x64-1.2.0.tgz --strip=1
+  aria2c https://github.com/microsoft/onnxruntime/releases/download/v1.4.0/onnxruntime-linux-x64-1.4.0.tgz
+  tar -zxf onnxruntime-linux-x64-1.4.0.tgz --strip=1
   cp -r include/* /usr/include 
   cp -r lib/* /usr/lib
   ldconfig /usr/lib
diff --git a/tools/ci_build/github/linux/docker/scripts/requirements.txt b/tools/ci_build/github/linux/docker/scripts/requirements.txt
index fc5f5e5175..be12b008b7 100644
--- a/tools/ci_build/github/linux/docker/scripts/requirements.txt
+++ b/tools/ci_build/github/linux/docker/scripts/requirements.txt
@@ -4,4 +4,5 @@ mypy
 pytest
 setuptools>=41.4.0
 wheel
-onnx==1.7.0
\ No newline at end of file
+onnx==1.7.0
+argparse
\ No newline at end of file