Change target tritonfastertransformerbackend to trtonturbomindbackend (…

…InternLM#36)

* change target tritonfastertransformerbackend to tritonturbomindbackend

* install targets to backends/turbomind

* changge model_dir

CMakeLists.txt

@@ -374,8 +374,8 @@ install(
     transformer-shared
   EXPORT
     transformer-shared-targets
-  LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/fastertransformer
-  ARCHIVE DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/fastertransformer
+  LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
+  ARCHIVE DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
 )
 
 install(

README.md

@@ -100,7 +100,7 @@ Run one of the following commands to serve a LLaMA model on NVIDIA GPU server:
 ```shell
 python3 lmdeploy/serve/turbomind/deploy.py llama-7B /path/to/llama-7b llama \
     --tokenizer_path /path/to/tokenizer/model
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -111,7 +111,7 @@ bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fast
 ```shell
 python3 lmdeploy/serve/turbomind/deploy.py llama-13B /path/to/llama-13b llama \
     --tokenizer_path /path/to/tokenizer/model --tp 2
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -129,7 +129,7 @@ python3 -m fastchat.model.apply_delta \
   --delta-path lmsys/vicuna-7b-delta-v1.1
 
 python3 lmdeploy/serve/turbomind/deploy.py vicuna-7B /path/to/vicuna-7b hf
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -145,7 +145,7 @@ python3 -m fastchat.model.apply_delta \
   --delta-path lmsys/vicuna-13b-delta-v1.1
 
 python3 lmdeploy/serve/turbomind/deploy.py vicuna-13B /path/to/vicuna-13b hf
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>

README_zh-CN.md

@@ -98,7 +98,7 @@ make -j$(nproc) && make install
 ```shell
 python3 lmdeploy/serve/turbomind/deploy.py llama-7B /path/to/llama-7b llama \
     --tokenizer_path /path/to/tokenizer/model
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -109,7 +109,7 @@ bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fast
 ```shell
 python3 lmdeploy/serve/turbomind/deploy.py llama-13B /path/to/llama-13b llama \
     --tokenizer_path /path/to/tokenizer/model --tp 2
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -127,7 +127,7 @@ python3 -m fastchat.model.apply_delta \
   --delta-path lmsys/vicuna-7b-delta-v1.1
 
 python3 lmdeploy/serve/turbomind/deploy.py vicuna-7B /path/to/vicuna-7b hf
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>
@@ -143,7 +143,7 @@ python3 -m fastchat.model.apply_delta \
   --delta-path lmsys/vicuna-13b-delta-v1.1
 
 python3 lmdeploy/serve/turbomind/deploy.py vicuna-13B /path/to/vicuna-13b hf
-bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/fastertransformer
+bash workspace/service_docker_up.sh --lib-dir $(pwd)/build/install/backends/turbomind
 ```
 
 </details>

examples/cpp/llama/llama_config.ini

@@ -2,8 +2,8 @@
 data_type=fp16
 enable_custom_all_reduce=0
 pipeline_para_size=1
-tensor_para_size=8
-model_dir=/shared_data/chatpjlm-0/v0.2.3/fastertransformer/weights/
+tensor_para_size=1
+model_dir=/workspace/models/triton_models/weights/
 
 
 [request]

lmdeploy/serve/turbomind/deploy.py

@@ -227,7 +227,7 @@ def get_param(_name, _size):
         del ckpt
 
     for name, param in model_params.items():
-        # transpose all weights as FasterTransformer is expecting column-major
+        # transpose all weights as TurboMind is expecting column-major
         # weights: (output_dims, input_dims) -> (input_dims, output_dims)
         key = name.split('.')[-2]
         if key in ['w1', 'w3', 'wq', 'wk', 'wv', 'w2', 'wo']:

lmdeploy/serve/turbomind/service_docker_up.sh

@@ -5,7 +5,7 @@ show_help() {
   echo
   echo "Options:"
   echo "  -h, --help   Show this help message and exit"
-  echo "  --lib-dir    Specify the directory of fastertransformer libraries"
+  echo "  --lib-dir    Specify the directory of turbomind libraries"
 }
 
 # check if '-h' or '--help' in the arguments
@@ -64,7 +64,7 @@ for ((i = 1; i <= $#; i++)); do
       docker run \
         --gpus $DEVICES \
         --rm \
-        -v "${LIB_PATH}":/opt/tritonserver/backends/fastertransformer \
+        -v "${LIB_PATH}":/opt/tritonserver/backends/turbomind \
         -v ""${SCRIPT_ABS_DIR}"":/workspace/models \
         --shm-size 16g \
         -p 33336:22 \

lmdeploy/serve/turbomind/triton_models/interactive/config.pbtxt

@@ -25,7 +25,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 name: "turbomind"
-backend: "fastertransformer"
+backend: "turbomind"
 default_model_filename: "weights"
 max_batch_size: 1
 

src/turbomind/models/llama/llama_utils.cu

@@ -148,7 +148,7 @@ size_t curandStateGetSize()
 bool isDebug()
 {
     static const bool is_debug = [] {
-        const auto level = std::getenv("FT_DEBUG_LEVEL");
+        const auto level = std::getenv("TM_DEBUG_LEVEL");
         if (level && level == std::string("DEBUG")) {
             return true;
         }

src/turbomind/triton_backend/CMakeLists.txt

@@ -26,7 +26,7 @@
 
 cmake_minimum_required (VERSION 3.18)
 
-project(tritonfastertransformerbackend LANGUAGES C CXX)
+project(tritonturbomindbackend LANGUAGES C CXX)
 
 #
 # Options
@@ -89,12 +89,12 @@ endif() # TRITON_ENABLE_GPU
 configure_file(libtriton_fastertransformer.ldscript libtriton_fastertransformer.ldscript COPYONLY)
 
 add_library(
-  triton-fastertransformer-backend SHARED
+  triton-turbomind-backend SHARED
   libfastertransformer.cc
 )
 
 add_library(
-  TritonFasterTransformerBackend::triton-fastertransformer-backend ALIAS triton-fastertransformer-backend
+  TritonTurboMindBackend::triton-turbomind-backend ALIAS triton-turbomind-backend
 )
 
 find_package(CUDAToolkit REQUIRED)
@@ -106,13 +106,13 @@ endif()
 
 set(CUDA_PATH ${CUDA_TOOLKIT_ROOT_DIR})
 
-target_compile_definitions(triton-fastertransformer-backend
+target_compile_definitions(triton-turbomind-backend
   PUBLIC
   USE_TRITONSERVER_DATATYPE
   BUILD_MULTI_GPU)
 
 target_include_directories(
-  triton-fastertransformer-backend
+  triton-turbomind-backend
   PRIVATE
   ${CMAKE_CURRENT_SOURCE_DIR}/src
   ${TRITON_PYTORCH_INCLUDE_PATHS}
@@ -123,31 +123,31 @@ target_include_directories(
   )
 
 target_link_directories(
-  triton-fastertransformer-backend
+  triton-turbomind-backend
   PRIVATE
   ${CUDA_PATH}/lib64
   )
 
-target_compile_features(triton-fastertransformer-backend PRIVATE cxx_std_14)
+target_compile_features(triton-turbomind-backend PRIVATE cxx_std_14)
 
 target_compile_options(
-  triton-fastertransformer-backend PRIVATE
+  triton-turbomind-backend PRIVATE
   $<$<OR:$<CXX_COMPILER_ID:Clang>,$<CXX_COMPILER_ID:AppleClang>,$<CXX_COMPILER_ID:GNU>>:
     -Wall -Wextra -Wno-unused-parameter -Wno-type-limits >#-Werror>
 )
 
 if(${TRITON_ENABLE_GPU})
   target_compile_definitions(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PRIVATE TRITON_ENABLE_GPU=1
   )
 endif() # TRITON_ENABLE_GPU
 
 set_target_properties(
-  triton-fastertransformer-backend
+  triton-turbomind-backend
   PROPERTIES
     POSITION_INDEPENDENT_CODE ON
-    OUTPUT_NAME triton_fastertransformer
+    OUTPUT_NAME triton_turbomind
     SKIP_BUILD_RPATH TRUE
     BUILD_WITH_INSTALL_RPATH TRUE
     INSTALL_RPATH_USE_LINK_PATH FALSE
@@ -159,7 +159,7 @@ set_target_properties(
 # Need to turn off unused-but-set-variable due to Torchvision
 # Need to turn off unknown-pragmas due to ATen OpenMP
 set_target_properties(
-  triton-fastertransformer-backend
+  triton-turbomind-backend
   PROPERTIES COMPILE_FLAGS
     "-Wno-unknown-pragmas -Wno-unused-but-set-variable"
 )
@@ -170,7 +170,7 @@ FOREACH(p ${TRITON_PYTORCH_LIB_PATHS})
 ENDFOREACH(p)
 
 target_link_libraries(
-  triton-fastertransformer-backend
+  triton-turbomind-backend
   PRIVATE
     triton-core-serverapi  # from repo-core
     triton-core-backendapi # from repo-core
@@ -186,23 +186,23 @@ target_link_libraries(
 
 if (BUILD_MULTI_GPU)
   target_compile_definitions(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PUBLIC
       BUILD_MULTI_GPU
   )
   target_include_directories(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PRIVATE
       ${MPI_INCLUDE_PATH}
   )
   target_link_directories(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PRIVATE
       ${MPI_Libraries}
       /usr/local/mpi/lib
   )
   target_link_libraries(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PRIVATE
       ${NCCL_LIBRARIES}
       ${MPI_LIBRARIES}
@@ -211,7 +211,7 @@ endif()
 
 if(${TRITON_ENABLE_GPU})
   target_link_libraries(
-    triton-fastertransformer-backend
+    triton-turbomind-backend
     PRIVATE
       CUDA::cudart
   )
@@ -221,51 +221,51 @@ endif() # TRITON_ENABLE_GPU
 # Install
 #
 include(GNUInstallDirs)
-set(INSTALL_CONFIGDIR ${CMAKE_INSTALL_LIBDIR}/cmake/TritonFasterTransformerBackend)
+set(INSTALL_CONFIGDIR ${CMAKE_INSTALL_LIBDIR}/cmake/TurboMindBackend)
 
 install(
   TARGETS
-    triton-fastertransformer-backend
+    triton-turbomind-backend
   EXPORT
-    triton-fastertransformer-backend-targets
-  LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/fastertransformer
-  ARCHIVE DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/fastertransformer
+    triton-turbomind-backend-targets
+  LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
+  ARCHIVE DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
 )
 
 install(
   EXPORT
-    triton-fastertransformer-backend-targets
+    triton-turbomind-backend-targets
   FILE
-    TritonFasterTransformerBackendTargets.cmake
+    TritonTurboMindBackendTargets.cmake
   NAMESPACE
-    TritonFasterTransformerBackend::
+    TritonTurboMindBackend::
   DESTINATION
     ${INSTALL_CONFIGDIR}
 )
 
 include(CMakePackageConfigHelpers)
 configure_package_config_file(
-  ${CMAKE_SOURCE_DIR}/cmake/TritonFasterTransformerBackendConfig.cmake.in
-  ${CMAKE_CURRENT_BINARY_DIR}/TritonFasterTransformerBackendConfig.cmake
+  ${CMAKE_SOURCE_DIR}/cmake/TritonTurboMindBackendConfig.cmake.in
+  ${CMAKE_CURRENT_BINARY_DIR}/TritonTurboMindBackendConfig.cmake
   INSTALL_DESTINATION ${INSTALL_CONFIGDIR}
 )
 
 install(
   FILES
-  ${CMAKE_CURRENT_BINARY_DIR}/TritonFasterTransformerBackendConfig.cmake
+  ${CMAKE_CURRENT_BINARY_DIR}/TritonTurboMindBackendConfig.cmake
   DESTINATION ${INSTALL_CONFIGDIR}
 )
 
 #
 # Export from build tree
 #
 export(
-  EXPORT triton-fastertransformer-backend-targets
-  FILE ${CMAKE_CURRENT_BINARY_DIR}/TritonFasterTransformerBackendTargets.cmake
-  NAMESPACE TritonFasterTransformerBackend::
+  EXPORT triton-turbomind-backend-targets
+  FILE ${CMAKE_CURRENT_BINARY_DIR}/TritonTurboMindBackendTargets.cmake
+  NAMESPACE TritonTurboMindBackend::
 )
 
-export(PACKAGE TritonFasterTransformerBackend)
+export(PACKAGE TritonTurboMindBackend)
 
 
 # Copyright (c) 2021-2023, NVIDIA CORPORATION.  All rights reserved.

src/turbomind/triton_backend/libfastertransformer.cc

@@ -511,11 +511,11 @@ TRITONSERVER_Error* ModelState::AutoCompleteConfig()
         }
     }
     else {
-        // Auto-complete configuration is not supported since fastertransformer does
+        // Auto-complete configuration is not supported since turbomind does
         // not store/capture sufficient model metadata so just log error instead.
         LOG_MESSAGE(TRITONSERVER_LOG_WARN,
                     (std::string("skipping model configuration auto-complete for '") + Name()
-                     + "': not supported for fastertransformer backend")
+                     + "': not supported for turbomind backend")
                         .c_str());
     }
 
@@ -940,7 +940,7 @@ void ModelInstanceState::ProcessRequests(TRITONBACKEND_Request** requests, const
                 request_count,
                 TRITONSERVER_ErrorNew(
                     TRITONSERVER_ERROR_INTERNAL,
-                    std::string("null request given to FasterTransformer backend for '" + Name() + "'").c_str()));
+                    std::string("null request given to TurboMind backend for '" + Name() + "'").c_str()));
             return;
         }
 
@@ -1115,7 +1115,7 @@ void ModelInstanceState::ProcessRequests(TRITONBACKEND_Request** requests, const
     for (auto& response : responses) {
         if (response != nullptr) {
             LOG_IF_ERROR(TRITONBACKEND_ResponseSend(response, TRITONSERVER_RESPONSE_COMPLETE_FINAL, nullptr),
-                         "failed to send FasterTransformer backend response");
+                         "failed to send TurboMind backend response");
             LOG_MESSAGE(TRITONSERVER_LOG_VERBOSE, (std::string("response is sent")).c_str());
         }
         else {
@@ -1160,7 +1160,7 @@ void streaming_callback(std::shared_ptr<std::unordered_map<std::string, Tensor>>
         if (response != nullptr) {
             LOG_MESSAGE(TRITONSERVER_LOG_VERBOSE, (std::string("start to send streaming response")).c_str());
             LOG_IF_ERROR(TRITONBACKEND_ResponseSend(response, 0, nullptr),
-                         "failed to send FasterTransformer backend response");
+                         "failed to send TurboMind backend response");
             LOG_MESSAGE(TRITONSERVER_LOG_VERBOSE, (std::string("streaming response is sent")).c_str());
         }
         else {
@@ -1358,7 +1358,7 @@ ModelInstanceState::Execute(std::vector<TRITONBACKEND_Response*>*
             responses,
             response_count,
             TRITONSERVER_ErrorNew(TRITONSERVER_ERROR_INTERNAL,
-                                  ("FasterTransformer execute failure: " + std::string(ex.what())).c_str()));
+                                  ("TurboMind execute failure: " + std::string(ex.what())).c_str()));
     }
     auto output_tensors = output_tensors_list[0];
     return output_tensors;