Merge branch 'refactor-rm-offline-inference-minseokl' into 'main'

Remove the offline inference in the code level

See merge request dl/hugectr/hugectr!1466

.gitlab-ci.yml

@@ -157,7 +157,8 @@ build_inference:
     DST_IMAGE: $INFER_IMAGE_VERSIONED
     CMAKE_OPTION: "-DENABLE_INFERENCE=ON -DCMAKE_BUILD_TYPE=Release -DSM=\"70;75;80;90\" -DCLANGFORMAT=OFF"
     BUILD_HUGECTR: 1
-    BUILD_HUGECTR_BACKEND: 1
+    BUILD_HPS_BACKEND: 1
+    HUGECTR_BACKEND_VER: main
     TRITON_BRANCH: ${TARGET_TRITON_BRANCH}
 
 build_sok_tf2:
@@ -530,19 +531,6 @@ e2e_nvt_regression_test:
     DGXNNODES: 1
     TEST_CMD: ./ci/integration_test/nvt/nvt_regression_test.sub
 
-notebook_hugectr:
-  extends: .cluster_test_job_daily
-  needs:
-    - build_train_single_node
-  variables:
-    GPFSFOLDER: $LOGDIR/notebook_hugectr
-    GIT_CLONE_PATH: ${GIT_CLONE_PATH_SELENE}
-    CONT: $TRAIN_IMAGE_VERSIONED
-    MOUNTS: /lustre/fsw/devtech/hpc-hugectr/criteo_1TB/day_0:/workdir/tools/day_0,/lustre/fsw/devtech/hpc-hugectr/criteo_1TB/day_1:/workdir/tools/day_1
-    WALLTIME: "01:00:00"
-    DGXNNODES: 1
-    TEST_CMD: ./ci/integration_test/notebooks/notebook_hugectr.sub
-
 nb_hps_demo:
   extends: .cluster_test_job_daily
   needs:

CMakeLists.txt

@@ -117,6 +117,13 @@ find_package(OpenMP REQUIRED)
 find_package(Threads)
 
 option(ENABLE_MULTINODES "Enable multi-nodes training" OFF)
+option(ENABLE_INFERENCE "Enable Inference" OFF)
+
+if(ENABLE_MULTINODES AND ENABLE_INFERENCE)
+  message(WARNING "Inference can be only enabled with the multi-node mode off. Set ENABLE_MULTINODES=OFF")
+  set(ENABLE_MULTINODES OFF)
+endif()
+
 if(ENABLE_MULTINODES)
   message(STATUS "Multi Node Enabled")
   find_package(MPI)
@@ -141,7 +148,6 @@ if (KEY_HIT_RATIO)
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DKEY_HIT_RATIO")
 endif()
 
-option(ENABLE_INFERENCE "Enable Inference" OFF)
 if(ENABLE_INFERENCE)
   add_definitions(-DLIBCUDACXX_VERSION)
 endif()
@@ -324,11 +330,8 @@ if(ENABLE_INFERENCE)
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DENABLE_INFERENCE")
   add_subdirectory(HugeCTR/core23)
   add_subdirectory(HugeCTR/src/hps)
-  add_subdirectory(HugeCTR/src/inference)
   add_subdirectory(HugeCTR/src/inference_benchmark)
-  add_subdirectory(HugeCTR/src/cpu)
   add_subdirectory(test/utest/hps)
-  add_subdirectory(test/utest/inference)
 else()
   #setting binary files install path
   add_subdirectory(HugeCTR/src)

HugeCTR/include/core23_network.hpp

@@ -111,17 +111,6 @@ class Core23TempNetwork final {
    */
   void upload_params_to_device(const std::string& model_file);
 
-  /**
-   * Read parameters from model_file.
-   */
-  void upload_params_to_device_inference(const std::string& model_file);
-
-  /**
-   * Read non-trainable parameters from model_file, e.g., running mean and running variable for
-   * BatchNorm
-   */
-  void upload_non_trainable_params_to_device_inference(const std::string& model_file);
-
   /**
    * Writing parameters to cpu buffer.
    */