some bugfix

- uuid crash issue
- update lora implement
- set page size by param
- delete deprecated files

.gitmodules

@@ -1,3 +1,3 @@
 [submodule "third_party/from_source/cutlass"]
-	path = third_party/from_source/cutlass
-	url = https://github.com/NVIDIA/cutlass.git
+    path = third_party/from_source/cutlass
+    url = https://github.com/NVIDIA/cutlass.git

CMakeLists.txt

@@ -45,6 +45,7 @@ set(CONFIG_HOST_CPU_TYPE "X86" CACHE STRING "host cpu type, like X86, ARMV9, etc
 ##
 option(ENABLE_NV_STATIC_LIB "build with static lib of nvidia" OFF)
 option(USE_SYSTEM_NV_LIB "use system nccl lib instead download binary." OFF)
+option(BUILD_HIEDNN "build HIE-DNN from source" OFF)
 option(ENABLE_CUDA_PINNED_WEIGHT_LOAD "enable cuda pinned memory for load weight" OFF)
 option(ENABLE_SPAN_ATTENTION "enable build with span attention" ON)
 option(ENABLE_MULTINUMA "enable multinuma, if on cpu multinuma service will be compiled" OFF)
@@ -174,18 +175,14 @@ if(LOCK_CHECK)
 endif()
 
 if (ENABLE_CUDA)
-    include(cuda)
     list(APPEND ALLSPARK_DEFINITION "-DENABLE_CUDA")
-
-    if (ENABLE_CUSPARSELT)
-      set(ENABLE_CUSPARSELT ON)
-      list(APPEND ALLSPARK_DEFINITION "-DENABLE_CUSPARSELT")
-    endif()
-
+        if (ENABLE_CUSPARSELT)
+        list(APPEND ALLSPARK_DEFINITION "-DENABLE_CUSPARSELT")
+        endif()
+    include(cuda)
     if (ENABLE_SPARSE)
-       list(APPEND ALLSPARK_DEFINITION "-DENABLE_SPARSE")
+        list(APPEND ALLSPARK_DEFINITION "-DENABLE_SPARSE")
     endif()
-
     if (ENABLE_FP8)
             list(APPEND ALLSPARK_DEFINITION "-DENABLE_FP8")
     endif()

build.sh

@@ -8,6 +8,7 @@ with_platform="${AS_PLATFORM:-cuda}"
 cuda_version="${AS_CUDA_VERSION:-12.4}"
 cuda_sm="${AS_CUDA_SM:-80;86;90a}"
 NCCL_VERSION="${AS_NCCL_VERSION:-2.23.4}"
+build_folder="${AS_BUILD_FOLDER:-build}"
 
 ## NCCL Version Map:
 ## the corresponding pre-build nccl will download on oss.
@@ -22,9 +23,9 @@ build_type="${AS_BUILD_TYPE:-Release}"
 cuda_static="${AS_CUDA_STATIC:-OFF}"
 build_package="${AS_BUILD_PACKAGE:-ON}"
 enable_glibcxx11_abi="${AS_CXX11_ABI:-OFF}"
+build_hiednn="${AS_BUILD_HIEDNN:-ON}"
 enable_span_attn="${ENABLE_SPAN_ATTENTION:-ON}"
 enable_multinuma="${ENABLE_MULTINUMA:-OFF}"
-
 function clone_pull {
   GIT_URL=$1
   DIRECTORY=$2
@@ -42,11 +43,11 @@ function clone_pull {
 }
 
 if [ "$clean" == "ON" ]; then
-    rm -rf build
+    rm -rf ${build_folder}
 fi
 
-if [ ! -d "./build"  ]; then
-    mkdir build && cd build
+if [ ! -d "./${build_folder}"  ]; then
+    mkdir ${build_folder} && cd ${build_folder}
 
     conan profile new dashinfer_compiler_profile --detect --force
     conanfile=../conan/conanfile.txt
@@ -74,7 +75,7 @@ if [ ! -d "./build"  ]; then
     cd ../
 fi
 
-cd build
+cd ${build_folder}
 source ./activate.sh
 export PATH=`pwd`/bin:$PATH
 
@@ -94,6 +95,7 @@ if [ "${with_platform,,}" == "cuda" ]; then
       -DBUILD_PYTHON=OFF \
       -DALWAYS_READ_LOAD_MODEL=OFF \
       -DENABLE_SPAN_ATTENTION=${enable_span_attn} \
+      -DBUILD_HIEDNN=${build_hiednn} \
       -DENABLE_MULTINUMA=OFF
 elif [ "${with_platform,,}" == "x86" ]; then
   cmake .. \

cmake/cutlass.cmake

@@ -7,8 +7,8 @@ set(CUTLASS_NVCC_ARCHS ${CMAKE_CUDA_ARCHITECTURES} CACHE STRING "The SM architec
 set(CUTLASS_INSTALL ${INSTALL_LOCATION}/cutlass/install)
 message(STATUS "CUTLASS_INSTALL: ${CUTLASS_INSTALL}")
 
-message(STATUS "Use cutlass from submodule")
-set(CUTLASS_SOURCE_DIR ${PROJECT_SOURCE_DIR}/third_party/from_source/cutlass)
+  message(STATUS "Use cutlass from submodule")
+  set(CUTLASS_SOURCE_DIR ${PROJECT_SOURCE_DIR}/third_party/from_source/cutlass)
 
 include(ExternalProject)
 

cmake/flash-attention.cmake

@@ -35,12 +35,9 @@ else()
   set(FLASHATTN_LIBRARY_NAME libflash-attn.so)
 endif()
 
-option(FLASHATTN_BUILD_FROM_SOURCE "build flash-attn from source or use prebuilt lib" ON)
-# make sure you copy prebuild and source code under csrc together, to avoid header aligment issue.
 
 include(ExternalProject)
 
-if (FLASHATTN_BUILD_FROM_SOURCE)
   message(STATUS "build flash-attention from source")
 
     message(STATUS "Use flash-attention from external project")
@@ -78,9 +75,6 @@ if (FLASHATTN_BUILD_FROM_SOURCE)
   ExternalProject_Get_Property(project_flashattn SOURCE_SUBDIR)
   set(FLASHATTN_INCLUDE_DIR ${SOURCE_DIR}/${SOURCE_SUBDIR})
 
-else() # FLASHATTN_BUILD_FROM_SOURCE
-    message(FATAL_ERROR "flash attention build only support source code build.")
-endif() # FLASHATTN_BUILD_FROM_SOURCE
 
 message(STATUS "FLASHATTN_LIBRARY_PATH: ${FLASHATTN_LIBRARY_PATH}")
 message(STATUS "FLASHATTN_INCLUDE_DIR: ${FLASHATTN_INCLUDE_DIR}")

cmake/hie-dnn.cmake

@@ -29,7 +29,8 @@ set(HIEDNN_LIBRARY_PATH ${HIEDNN_INSTALL}/lib64/libhiednn_static.a)
 message(STATUS "HIEDNN_INSTALL: ${HIEDNN_INSTALL}")
 message(STATUS "HIEDNN_LIBRARY_PATH: ${HIEDNN_LIBRARY_PATH}")
 
-set(HIEDNN_SOURCE_DIR ${PROJECT_SOURCE_DIR}/HIE-DNN)
+  set(HIEDNN_SOURCE_DIR ${PROJECT_SOURCE_DIR}/HIE-DNN)
+message(STATUS "Build HIE-DNN in: ${HIEDNN_SOURCE_DIR}")
 
 include(ExternalProject)
 

csrc/CMakeLists.txt

@@ -41,7 +41,7 @@ set(ALLSPARK_INC_DIR
     ${CMAKE_BINARY_DIR}/csrc
 )
 
-list(APPEND ALLSPARK_INC_DIR ${PROJECT_SOURCE_DIR}/third_party/from_source/cutlass/include)
+    list(APPEND ALLSPARK_INC_DIR ${PROJECT_SOURCE_DIR}/third_party/from_source/cutlass/include)
 
 set(ALLSPARK_3RD_LIBS
     ${THREAD_LIB}