feat: Add Mobile CI.

.github/workflows/main.yml

@@ -15,7 +15,7 @@ permissions:
 jobs:
   gtest:
     name: Unit Tests
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
     steps:
       - name: Checkout
         uses: actions/[email protected]
@@ -31,7 +31,7 @@ jobs:
         with:
           python-version: '3.9'
           cache: 'pip' # caching pip dependencies
-      - run: cd test && pip install -r requirements.txt
+      - run: cd test && pip install -r requirements.txt && pip install cmake
       - name: Build
         run: |
           cmake -DTEST=on -Bbuild && cmake --build build --target all
@@ -51,6 +51,41 @@ jobs:
           post: |
             COMMIT_MESSAGE="$(git show -s --format=%s)" python3 ../tools/ci/report.py
 
-
+  mobile_test:
+    name: Unit Tests on Mobile
+    runs-on: self-hosted
+    steps:
+      - name: Checkout
+        uses: actions/[email protected]
+        with:
+          submodules: true
+      - name: Security Check
+        #        Check if is a forked repo, if so, do not run action job.
+        if: github.repository != 'UbiquitousLearning/mllm'
+        run: |
+          echo "This is a forked repo, stop running"
+          exit 1
+      - uses: actions/setup-python@v4
+        with:
+          python-version: '3.9'
+          cache: 'pip' # caching pip dependencies
+      - run: cd test && pip install -r requirements.txt && pip install cmake
+      - name: Prepare & Build
+        run: |
+          bash tools/ci/build_arm64.sh
+      - name: Test
+        uses: gacts/run-and-post-run@v1
+        env:
+          FEISHU_TOKEN: ${{ secrets.FEISHU_TOKEN }}
+          FEISHU_AT: ${{ secrets.FEISHU_AT }}
+          PR_NUMBER: ${{ github.event.number }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          IS_MOBILE: 1
+        with:
+          run: adb devices && python3 ./tools/ci/mobile_test.py
+          post: |
+            COMMIT_MESSAGE="$(git show -s --format=%s)" python3 ./tools/ci/report.py
+
+
 
 

.gitignore

@@ -5,6 +5,7 @@
 build*/
 build/
 bin/
+bin*/
 .clangd/
 compile_commands.json
 *.bin
@@ -19,4 +20,6 @@ __pycache__/
 _deps
 .uuid
 *.cmake
-Makefile
+Makefile
+*.model
+*.mllm

CMakeLists.txt

@@ -1,152 +1,169 @@
 cmake_minimum_required(VERSION 3.10)
 
 project(mllm)
-cmake_policy(SET CMP0074 NEW) 
+cmake_policy(SET CMP0074 NEW)
 set(CMAKE_CXX_STANDARD 17)
-if (ARM)
+option(ARM "build on ARM" OFF)
+
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
+    message(STATUS "ARM detected")
+    set(ARM ON)
+    set(ANDROID_PLATFORM android-28)
+
+    # add_compile_options(-std=c++17)
+endif()
+
+if(ARM)
     set(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/../bin-arm)
 else()
     set(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/../bin)
-endif ()
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+endif()
 
 # temp executable name
 set(TEST_EXE main_test)
 option(TEST "test mode" ON)
 option(QUANT "quantize tools" ON)
-if (CMAKE_VERSION VERSION_GREATER_EQUAL "3.24.0")
+
+if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.24.0")
     cmake_policy(SET CMP0135 NEW)
-endif ()
+endif()
+
 include(FetchContent)
 FetchContent_Declare(
-        googletest
-        URL https://github.com/google/googletest/archive/03597a01ee50ed33e9dfd640b249b4be3799d395.zip
+    googletest
+    URL https://github.com/google/googletest/archive/03597a01ee50ed33e9dfd640b249b4be3799d395.zip
 )
 
-
 # find_program(CLANG_TIDY_EXE NAMES "clang-tidy" REQUIRED)
-#set(CMAKE_CXX_CLANG_TIDY "clang-tidy")
-
+# set(CMAKE_CXX_CLANG_TIDY "clang-tidy")
 option(DEBUG "debug print" OFF)
+
 # backend options
 option(NNAPI "Enable NNAPI" OFF)
 
 if(CMAKE_HOST_UNIX)
     message("current platform: Linux ")
-#    set(CMAKE_CXX_FLAGS "${CMAKE_C_FLAGS} -std=c++14 -stdlib=libc++")
+
+# set(CMAKE_CXX_FLAGS "${CMAKE_C_FLAGS} -std=c++14 -stdlib=libc++")
 elseif(CMAKE_HOST_WIN32)
     message("current platform: Windows ")
 else()
-    message("current platform: unkonw ")
+    message("current platform: unknown ")
 endif()
 
-aux_source_directory(${PROJECT_SOURCE_DIR}/src  DIR_SRC)
+aux_source_directory(${PROJECT_SOURCE_DIR}/src DIR_SRC)
+
 # aux_source_directory(${PROJECT_SOURCE_DIR}/src/backends/cpu  DIR_SRC_CPU)
-aux_source_directory(${PROJECT_SOURCE_DIR}/src/express  DIR_SRC_EXP)
-#aux_source_directory(${PROJECT_SOURCE_DIR}/src/quantize  DIR_SRC_QUANT)
-aux_source_directory(${PROJECT_SOURCE_DIR}/src/memory  DIR_SRC_MEM_MANAGER)
-aux_source_directory(${PROJECT_SOURCE_DIR}/examples  EMP_SRC)
-aux_source_directory(${PROJECT_SOURCE_DIR}/test  TEST_SRC)
-aux_source_directory(${PROJECT_SOURCE_DIR}/src/quantizer QUANT_SRC)
+aux_source_directory(${PROJECT_SOURCE_DIR}/src/express DIR_SRC_EXP)
 
+# aux_source_directory(${PROJECT_SOURCE_DIR}/src/quantize  DIR_SRC_QUANT)
+aux_source_directory(${PROJECT_SOURCE_DIR}/src/memory DIR_SRC_MEM_MANAGER)
+aux_source_directory(${PROJECT_SOURCE_DIR}/examples EMP_SRC)
+aux_source_directory(${PROJECT_SOURCE_DIR}/test TEST_SRC)
+aux_source_directory(${PROJECT_SOURCE_DIR}/src/quantizer QUANT_SRC)
 
 include_directories(${PROJECT_SOURCE_DIR}/src)
 include_directories(${PROJECT_SOURCE_DIR}/include)
 include_directories(${PROJECT_SOURCE_DIR}/thirdpartys/include)
 
 add_subdirectory(${CMAKE_CURRENT_LIST_DIR}/src/backends/cpu)
+
 # if compile to x86_64
-if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(x86_64|i686|AMD64)$")
-message(STATUS "x86_64 detected")
-add_compile_options(-mf16c)
-add_compile_options(-mavx2)
-elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
+if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(x86_64|i686|AMD64)$")
+    message(STATUS "x86_64 detected")
+    add_compile_options(-mf16c)
+    add_compile_options(-mavx2)
+elseif(${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
     message(STATUS "ARM detected")
 endif()
-if (QUANT)
+
+if(QUANT)
     include_directories(${PROJECT_SOURCE_DIR}/src/quantizer)
     file(GLOB_RECURSE MLLM_QUANT
 
-            ${PROJECT_SOURCE_DIR}/src/backends/cpu/quantize/*.hpp
-            ${PROJECT_SOURCE_DIR}/src/backends/cpu/quantize/*.cpp
+        ${PROJECT_SOURCE_DIR}/src/backends/cpu/quantize/*.hpp
+        ${PROJECT_SOURCE_DIR}/src/backends/cpu/quantize/*.cpp
     )
-    # MLLM Quantizer is /src/quantizer/*.cpp but except main.cpp
 
+    # MLLM Quantizer is /src/quantizer/*.cpp but except main.cpp
     file(GLOB_RECURSE MLLM_QUANTIZER
-            ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/*.cpp
-            ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/*.hpp)
-list(REMOVE_ITEM MLLM_QUANTIZER ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/main.cpp)
+        ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/*.cpp
+        ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/*.hpp)
+    list(REMOVE_ITEM MLLM_QUANTIZER ${CMAKE_CURRENT_LIST_DIR}/src/quantizer/main.cpp)
     message(STATUS "MLLM_Quant: ${MLLM_QUANT}")
     add_executable(
-            quantize
-            ${PROJECT_SOURCE_DIR}/src/quantizer/main.cpp
-            ${MLLM_QUANT}
-            ${MLLM_QUANTIZER}
-            #            ${DIR_SRC}
-            ${PROJECT_SOURCE_DIR}/src/ParamLoader.cpp
+        quantize
+        ${PROJECT_SOURCE_DIR}/src/quantizer/main.cpp
+        ${MLLM_QUANT}
+        ${MLLM_QUANTIZER}
 
+        # ${DIR_SRC}
+        ${PROJECT_SOURCE_DIR}/src/ParamLoader.cpp
     )
+endif()
 
-
-endif ()
-
-
-if (TEST)
+if(TEST)
     add_definitions(-DTEST)
     FetchContent_MakeAvailable(googletest)
     enable_testing()
     include_directories(${PROJECT_SOURCE_DIR}/test)
+
     # MLLM Tests
     file(GLOB_RECURSE MLLM_TEST
-            ${CMAKE_CURRENT_LIST_DIR}/test/*.cpp
-            ${CMAKE_CURRENT_LIST_DIR}/test/*.hpp
+        ${CMAKE_CURRENT_LIST_DIR}/test/*.cpp
+        ${CMAKE_CURRENT_LIST_DIR}/test/*.hpp
     )
     list(REMOVE_ITEM MLLM_TEST ${CMAKE_CURRENT_LIST_DIR}/test/TestSystemMemoryManager.cpp)
     list(REMOVE_ITEM MLLM_TEST ${CMAKE_CURRENT_LIST_DIR}/test/TestMemoryPoolManager.cpp)
 
-
     message(STATUS "MLLM_TEST: ${MLLM_TEST}")
+
     # Add ${DIR_SRC_QUANT} to ${DIR_SRC}
     # if ()
     add_executable(
-            MLLM_TEST
-            ${PROJECT_SOURCE_DIR}/test/main.cpp
-            ${MLLM_TEST}
-            ${DIR_SRC_CPU} ${DIR_SRC_MEM_MANAGER} ${DIR_SRC_EXP} ${DIR_SRC} ${MLLM_QUANTIZER}
+        MLLM_TEST
+        ${PROJECT_SOURCE_DIR}/test/main.cpp
+        ${MLLM_TEST}
+        ${DIR_SRC_CPU} ${DIR_SRC_MEM_MANAGER} ${DIR_SRC_EXP} ${DIR_SRC} ${MLLM_QUANTIZER}
     )
 
     target_link_libraries(
-            MLLM_TEST
-            GTest::gtest_main
-            MLLM_CPU
+        MLLM_TEST
+        GTest::gtest_main
+        MLLM_CPU
     )
     add_executable(
         memoryPoolTest
         ${PROJECT_SOURCE_DIR}/test/TestMemoryPoolManager.cpp
-        ${DIR_SRC_MEM_MANAGER}  ${PROJECT_SOURCE_DIR}/src/MemoryManager.hpp
+        ${DIR_SRC_MEM_MANAGER} ${PROJECT_SOURCE_DIR}/src/MemoryManager.hpp
     )
     include(GoogleTest)
-    gtest_discover_tests(MLLM_TEST)
-endif ()
-
-#else ()
-#    No Unit Test.
-add_executable(${TEST_EXE} ${PROJECT_SOURCE_DIR}/demo/main_test.cpp ${DIR_SRC_CPU} ${DIR_SRC_MEM_MANAGER} ${DIR_SRC_EXP} ${DIR_SRC} #${DIR_SRC_QUANT}
-        src/tokenizers/Tokenizer.cpp
-        src/tokenizers/Tokenizer.hpp
-        src/tokenizers/BPE/Bpe.cpp
-        src/tokenizers/BPE/Bpe.hpp)
-target_link_libraries(${TEST_EXE} MLLM_CPU )
+
+    # if not cross-compiling
+    if(NOT ARM)
+        gtest_discover_tests(MLLM_TEST)
+    endif()
+endif()
+
+# else ()
+# No Unit Test.
+add_executable(${TEST_EXE} ${PROJECT_SOURCE_DIR}/demo/main_test.cpp ${DIR_SRC_CPU} ${DIR_SRC_MEM_MANAGER} ${DIR_SRC_EXP} ${DIR_SRC} # ${DIR_SRC_QUANT}
+    src/tokenizers/Tokenizer.cpp
+    src/tokenizers/Tokenizer.hpp
+    src/tokenizers/BPE/Bpe.cpp
+    src/tokenizers/BPE/Bpe.hpp)
+target_link_libraries(${TEST_EXE} MLLM_CPU)
 
 # nnapi temp test
 if(NNAPI)
-#    include(/Users/luis/Library/Android/sdk/ndk/25.1.8937393/build/cmake/android.toolchain.cmake)
     include(${ANDROID_NDK}/build/cmake/android.toolchain.cmake)
     add_definitions(-DNNAPI_ENABLED)
+
     # TODO: better debug log
     add_definitions(-DDEBUG)
     add_subdirectory(${CMAKE_CURRENT_LIST_DIR}/src/backends/nnapi)
-    add_executable(nnapi_test ${PROJECT_SOURCE_DIR}/demo/nnapi_test.cpp ${DIR_SRC_CPU} ${DIR_SRC_EXP} ${DIR_SRC} )#${DIR_SRC_QUANT})
+    add_executable(nnapi_test ${PROJECT_SOURCE_DIR}/demo/nnapi_test.cpp ${DIR_SRC_CPU} ${DIR_SRC_EXP} ${DIR_SRC}) # ${DIR_SRC_QUANT})
     target_link_libraries(nnapi_test MLLM_CPU MLLM_NNAPI)
 endif()
-

src/Op.hpp

@@ -91,7 +91,7 @@ class Op {
         return NO_ERROR;
     }
 
-    virtual ErrorCode load(ParamLoader &loader) {
+    virtual ErrorCode load(AbstructLoader &loader) {
         // check inputs shape
         // reshape outputs
 #ifdef DEBUG

src/ParamLoader.hpp

@@ -43,8 +43,15 @@ static std::string readString(FILE *fp_) {
     delete[] tmp;
     return str;
 }
+
 #define _MAGIC_NUMBER 20012
-class ParamLoader {
+class AbstructLoader {
+public:
+    virtual bool load(mllm::Tensor *tensor) = 0;
+    virtual bool load(std::shared_ptr<mllm::Tensor> tensor) = 0;
+    virtual DataType getDataType(string name) = 0;
+};
+class ParamLoader : public AbstructLoader {
     friend class QuantWriter;
 
 public:
@@ -53,11 +60,11 @@ class ParamLoader {
     ParamLoader(void *buffer);
 #endif
     ~ParamLoader();
-    bool load(mllm::Tensor *tensor);
-    bool load(std::shared_ptr<mllm::Tensor> tensor);
+    bool load(mllm::Tensor *tensor) override;
+    bool load(std::shared_ptr<mllm::Tensor> tensor) override;
     vector<std::string> getParamNames();
     std::tuple<uint8_t *, uint64_t> load(string name);
-    DataType getDataType(string name);
+    DataType getDataType(string name) override;
 
 private:
     FILE *fp_;

src/backends/cpu/CPUAttention.cpp

@@ -258,7 +258,7 @@ ErrorCode CPUAttention::execute(vector<shared_ptr<Tensor>> inputs, vector<shared
 //    outputs[0]->printData<float>();
     return NO_ERROR;
 }
-ErrorCode CPUAttention::load(ParamLoader &loader) {
+ErrorCode CPUAttention::load(AbstructLoader &loader) {
     Q_proj_->load(loader);
     K_proj_->load(loader);
     V_proj_->load(loader);

src/backends/cpu/CPUAttention.hpp

@@ -23,7 +23,7 @@ class CPUAttention final : public Op {
     virtual ~CPUAttention() = default;
     virtual ErrorCode reshape(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
     virtual ErrorCode setUp(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
-    virtual ErrorCode load(ParamLoader &loader) override;
+    virtual ErrorCode load(AbstructLoader &loader) override;
     virtual ErrorCode execute(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
     virtual ErrorCode free(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
 

src/backends/cpu/CPUEmbedding.cpp

@@ -1,4 +1,5 @@
 #include "CPUEmbedding.hpp"
+#include "ParamLoader.hpp"
 mllm::CPUEmbedding::CPUEmbedding(mllm::Backend *bn,  string opName, int hiddenSize, int vocabSize) :
     Op(bn, opName), hiddenSize_(hiddenSize), vocabSize_(vocabSize) {
     CHECK_GT(hiddenSize_, 0);
@@ -18,7 +19,7 @@ ErrorCode mllm::CPUEmbedding::reshape(vector<shared_ptr<Tensor>> inputs, vector<
     return Op::reshape(inputs, outputs);
 }
 
-ErrorCode mllm::CPUEmbedding::load(mllm::ParamLoader &loader) {
+ErrorCode mllm::CPUEmbedding::load(mllm::AbstructLoader &loader) {
     weight_.setName(name() + ".weight");
     weight_.reshape(1, 1, vocabSize_, hiddenSize_);
     weight_.setDtype(loader.getDataType(weight_.name()));

src/backends/cpu/CPUEmbedding.hpp

@@ -9,7 +9,7 @@ class CPUEmbedding final : public Op {
     explicit CPUEmbedding(Backend *bn, string opName, int hiddenSize, int vocabSize);
     ~CPUEmbedding() override = default;
     ErrorCode reshape(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
-    ErrorCode load(ParamLoader &loader) override;
+    ErrorCode load(AbstructLoader &loader) override;
     ErrorCode free(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;
     ErrorCode execute(vector<shared_ptr<Tensor>> inputs, vector<shared_ptr<Tensor>> outputs) override;