Add pytorch-labs/tokenizers into ET submodules

.ci/scripts/build_llama_android.sh

@@ -42,6 +42,10 @@ install_executorch_and_backend_lib() {
 
 build_llama_runner() {
     echo "Building llama runner for Android..."
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     ANDROID_ABI=arm64-v8a
     cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \

.ci/scripts/test_ane_static_llama.sh

@@ -17,6 +17,12 @@ fi
 
 which "${PYTHON_EXECUTABLE}"
 
+# Update tokenizers submodule
+pushd $EXECUTORCH_ROOT/extension/llm/tokenizers
+echo "Update tokenizers submodule"
+git submodule update --init
+popd
+
 pushd $EXECUTORCH_ROOT/examples/apple/coreml/llama
 
 # Download stories llama110m artifacts

.ci/scripts/test_llama.sh

@@ -173,6 +173,10 @@ cmake_install_executorch_libraries() {
 
 cmake_build_llama_runner() {
     echo "Building llama runner"
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     dir="examples/models/llama"
     retry cmake \
         -DCMAKE_INSTALL_PREFIX=cmake-out \

.gitmodules

@@ -28,15 +28,9 @@
 [submodule "backends/xnnpack/third-party/pthreadpool"]
 	path = backends/xnnpack/third-party/pthreadpool
 	url = https://github.com/Maratyszcza/pthreadpool.git
-[submodule "extension/llm/third-party/abseil-cpp"]
-	path = extension/llm/third-party/abseil-cpp
-	url = https://github.com/abseil/abseil-cpp.git
-[submodule "extension/llm/third-party/re2"]
-	path = extension/llm/third-party/re2
-	url = https://github.com/google/re2.git
-[submodule "extension/llm/third-party/sentencepiece"]
-	path = extension/llm/third-party/sentencepiece
-	url = https://github.com/google/sentencepiece.git
+[submodule "extension/llm/tokenizers"]
+	path = extension/llm/tokenizers
+	url = https://github.com/pytorch-labs/tokenizers.git
 [submodule "kernels/optimized/third-party/eigen"]
 	path = kernels/optimized/third-party/eigen
 	url = https://gitlab.com/libeigen/eigen.git

backends/qualcomm/scripts/build.sh

@@ -144,6 +144,10 @@ if [ "$BUILD_X86_64" = true ]; then
    EXAMPLE_ROOT=examples/qualcomm
    CMAKE_PREFIX_PATH="${BUILD_ROOT}/lib/cmake/ExecuTorch;${BUILD_ROOT}/third-party/gflags;"
 
+   echo "Update tokenizers submodule..."
+   pushd $PRJ_ROOT/extension/llm/tokenizers
+   git submodule update --init
+   popd
    cmake $PRJ_ROOT/$EXAMPLE_ROOT \
        -DCMAKE_BUILD_TYPE=$BUILD_TYPE \
        -DCMAKE_PREFIX_PATH=$CMAKE_PREFIX_PATH \

build/build_android_library.sh

@@ -70,6 +70,11 @@ build_android_native_library() {
   fi
   cmake --build "${CMAKE_OUT}" -j "${CMAKE_JOBS}" --target install --config "${EXECUTORCH_CMAKE_BUILD_TYPE}"
 
+  # Update tokenizers submodule
+  pushd extension/llm/tokenizers
+  echo "Update tokenizers submodule"
+  git submodule update --init
+  popd
   cmake extension/android \
     -DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK}/build/cmake/android.toolchain.cmake \
     -DANDROID_ABI="${ANDROID_ABI}" \

examples/mediatek/CMakeLists.txt

@@ -122,17 +122,17 @@ if(${ANDROID})
   )
   # Build ABSL and RE2
   set(EXTENSIONS_LLM_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm)
-  set(THIRD_PARTY_ABSL_DIR ${EXTENSIONS_LLM_DIR}/third-party/abseil-cpp)
-  set(THIRD_PARTY_RE2_DIR ${EXTENSIONS_LLM_DIR}/third-party/re2)
+  set(THIRD_PARTY_ABSL_DIR ${EXTENSIONS_LLM_DIR}/tokenizers/third-party/abseil-cpp)
+  set(THIRD_PARTY_RE2_DIR ${EXTENSIONS_LLM_DIR}/tokenizers/third-party/re2)
   set(ABSL_ENABLE_INSTALL ON)
   set(ABSL_PROPAGATE_CXX_STD ON)
   set(_pic_flag ${CMAKE_POSITION_INDEPENDENT_CODE})
   set(CMAKE_POSITION_INDEPENDENT_CODE ON)
   add_subdirectory(
-    ${THIRD_PARTY_ABSL_DIR} ${CMAKE_CURRENT_BINARY_DIR}/third-party/abseil
+    ${THIRD_PARTY_ABSL_DIR} ${CMAKE_CURRENT_BINARY_DIR}/tokenizers/third-party/abseil
   )
   add_subdirectory(
-    ${THIRD_PARTY_RE2_DIR} ${CMAKE_CURRENT_BINARY_DIR}/third-party/re2
+    ${THIRD_PARTY_RE2_DIR} ${CMAKE_CURRENT_BINARY_DIR}/tokenizers/third-party/re2
   )
   set(CMAKE_POSITION_INDEPENDENT_CODE ${_pic_flag})
 

examples/models/llama/runner/CMakeLists.txt

@@ -66,11 +66,11 @@ set(ABSL_PROPAGATE_CXX_STD ON)
 set(_pic_flag ${CMAKE_POSITION_INDEPENDENT_CODE})
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 add_subdirectory(
-  ${EXECUTORCH_ROOT}/extension/llm/third-party/abseil-cpp
+  ${EXECUTORCH_ROOT}/extension/llm/tokenizers/third-party/abseil-cpp
   ${CMAKE_CURRENT_BINARY_DIR}/abseil-cpp
 )
 add_subdirectory(
-  ${EXECUTORCH_ROOT}/extension/llm/third-party/re2
+  ${EXECUTORCH_ROOT}/extension/llm/tokenizers/third-party/re2
   ${CMAKE_CURRENT_BINARY_DIR}/re2
 )
 set(CMAKE_POSITION_INDEPENDENT_CODE ${_pic_flag})
@@ -82,6 +82,8 @@ set(llama_runner_deps executorch extension_data_loader extension_module
 target_link_libraries(llama_runner PUBLIC ${llama_runner_deps})
 
 target_include_directories(
-  llama_runner INTERFACE ${_common_include_directories} ${EXECUTORCH_ROOT}
+  llama_runner
+  INTERFACE ${_common_include_directories} ${EXECUTORCH_ROOT}
+            ${EXECUTORCH_ROOT}/extension/llm/tokenizers/include
 )
 target_compile_options(llama_runner PUBLIC ${_preprocessor_flag})

examples/models/llama/tokenizer/test/CMakeLists.txt

@@ -19,7 +19,7 @@ include(${EXECUTORCH_ROOT}/build/Test.cmake)
 
 set(_tokenizer_test_srcs
     test_tiktoken.cpp
-    ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizer/tiktoken.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizers/src/tiktoken.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/../llama_tiktoken.cpp
 )
 
@@ -29,11 +29,11 @@ set(ABSL_PROPAGATE_CXX_STD ON)
 set(_pic_flag ${CMAKE_POSITION_INDEPENDENT_CODE})
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/third-party/abseil-cpp
+  ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizers/third-party/abseil-cpp
   ${CMAKE_CURRENT_BINARY_DIR}/abseil-cpp
 )
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/third-party/re2
+  ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizers/third-party/re2
   ${CMAKE_CURRENT_BINARY_DIR}/re2
 )
 set(CMAKE_POSITION_INDEPENDENT_CODE ${_pic_flag})
@@ -43,5 +43,6 @@ target_include_directories(
   tokenizer_test
   PRIVATE
     ${CMAKE_INSTALL_PREFIX}/include
-    ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/third-party/abseil-cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizers/include
+    ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../extension/llm/tokenizers/third-party/abseil-cpp
 )

examples/qualcomm/CMakeLists.txt

@@ -63,7 +63,10 @@ target_compile_options(
   full_portable_ops_lib INTERFACE -DET_EVENT_TRACER_ENABLED
 )
 target_include_directories(
-  full_portable_ops_lib PUBLIC ${_common_include_directories}
+  full_portable_ops_lib
+  PUBLIC
+    ${_common_include_directories}
+    ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm/tokenizers/include
 )
 
 # find RE2 for tokenizer
@@ -72,11 +75,11 @@ set(ABSL_PROPAGATE_CXX_STD ON)
 set(_pic_flag ${CMAKE_POSITION_INDEPENDENT_CODE})
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm/third-party/abseil-cpp
+  ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm/tokenizers/third-party/abseil-cpp
   ${CMAKE_CURRENT_BINARY_DIR}/abseil-cpp
 )
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm/third-party/re2
+  ${CMAKE_CURRENT_SOURCE_DIR}/../../extension/llm/tokenizers/third-party/re2
   ${CMAKE_CURRENT_BINARY_DIR}/re2
 )
 set(CMAKE_POSITION_INDEPENDENT_CODE ${_pic_flag})

extension/llm/tokenizer/CMakeLists.txt

@@ -21,11 +21,11 @@ set(ABSL_PROPAGATE_CXX_STD ON)
 set(_pic_flag ${CMAKE_POSITION_INDEPENDENT_CODE})
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../third-party/abseil-cpp
+  ${CMAKE_CURRENT_SOURCE_DIR}/../tokenizers/third-party/abseil-cpp
   ${CMAKE_CURRENT_BINARY_DIR}/abseil-cpp
 )
 add_subdirectory(
-  ${CMAKE_CURRENT_SOURCE_DIR}/../third-party/re2
+  ${CMAKE_CURRENT_SOURCE_DIR}/../tokenizers/third-party/re2
   ${CMAKE_CURRENT_BINARY_DIR}/re2
 )
 set(CMAKE_POSITION_INDEPENDENT_CODE ${_pic_flag})
@@ -35,6 +35,7 @@ add_library(extension_llm_tokenizer ${_extension_llm_tokenizer__srcs})
 target_include_directories(
   extension_llm_tokenizer PUBLIC ${EXECUTORCH_ROOT}/..
                                  ${_common_include_directories}
+                                 ${CMAKE_CURRENT_SOURCE_DIR}/../tokenizers/include
 )
 
 target_link_libraries(extension_llm_tokenizer re2::re2)
@@ -53,7 +54,7 @@ install(
 target_include_directories(
   extension_llm_tokenizer
   PRIVATE ${CMAKE_INSTALL_PREFIX}/include
-          ${CMAKE_CURRENT_SOURCE_DIR}/../third-party/abseil-cpp
+          ${CMAKE_CURRENT_SOURCE_DIR}/../tokenizers/third-party/abseil-cpp
 )
 
 if(BUILD_TESTING)

shim_et/xplat/executorch/build/env_interface.bzl

@@ -43,9 +43,10 @@ _EXTERNAL_DEPS = {
     "nlohmann_json": [], # Intentionally not supporting OSS buck build HF tokenizer.
     "prettytable": "//third-party:prettytable",
     "pybind11": "//third-party:pybind11",
-    "re2": "//extension/llm/third-party:re2",
+    "re2": "//extension/llm/tokenizers/third-party:re2",
     "sentencepiece": [], # Intentionally not supporting OSS buck build of sentencepiece.
     "sentencepiece-py": [],
+    "tiktoken": "//extension/llm/tokenizers:tiktoken",
     # Core C++ PyTorch functionality like Tensor and ScalarType.
     "torch-core-cpp": "//third-party:libtorch",
     "torchgen": "//third-party:torchgen",