Merge branch 'main' into distribute-jobs-to-private-pools

Author: huydhn

.ci/docker/ci_commit_pins/optimum-executorch.txt

@@ -1 +1 @@
-a3942627f5ac048e06b4b1d703b0a6a53bf6da5b
+eea657ddbdeb1118943a92fb73c289985c3ee1ba

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-ab43fe4bdf5ccd82897f0e982c451a0127bd175e
+6fc0ad22f0a07b6f38d138861c56a765d5a9bb02

.ci/docker/common/install_zephyr.sh

@@ -8,6 +8,9 @@
 
 set -ex
 
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
 # Double check if the NDK version is set
 [ -n "${ZEPHYR_SDK}" ]
 
@@ -77,16 +80,8 @@ install_prerequiresites() {
         chmod +x kitware-archive.sh && \
         ./kitware-archive.sh && \
         rm -f kitware-archive.sh
-    useradd -d /home/zephyruser -m -s /bin/bash zephyruser
-}
-
-install_sdk() {
-    wget https://github.com/zephyrproject-rtos/sdk-ng/releases/download/v0.16.0/zephyr-sdk-0.16.0_linux-x86_64.tar.xz && \
-         tar -xf zephyr-sdk-0.16.0_linux-x86_64.tar.xz && \
-         rm -f zephyr-sdk-0.16.0_linux-x86_64.tar.xz && \
-         cd zephyr-sdk-0.16.0/ && \
-         ./setup.sh -c -t arm-zephyr-eabi
+    pip_install --no-cache-dir west
+    pip_install pyelftools
 }
 
 install_prerequiresites
-install_sdk

.ci/docker/ubuntu/Dockerfile

@@ -86,8 +86,9 @@ ARG ARM_SDK
 
 ARG ZEPHYR_SDK
 COPY ./common/install_zephyr.sh install_zephyr.sh
+COPY ./common/utils.sh utils.sh
 RUN if [ -n "${ZEPHYR_SDK}" ]; then bash ./install_zephyr.sh; fi
-RUN rm install_zephyr.sh
+RUN rm install_zephyr.sh utils.sh
 
 ARG QNN_SDK
 

.ci/scripts/build-qnn-sdk.sh

@@ -33,6 +33,8 @@ set_up_aot() {
       -DEXECUTORCH_BUILD_DEVTOOLS=ON \
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
       -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM_RUNNER=ON \
       -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \

.ci/scripts/setup-emscripten.sh

@@ -7,6 +7,13 @@
 
 set -ex
 
+# need version >= 17
+install_node() {
+    curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
+    source "$HOME/.nvm/nvm.sh"
+    nvm install 22
+}
+
 install_emscripten() {
     git clone https://github.com/emscripten-core/emsdk.git
     pushd emsdk || return
@@ -16,4 +23,5 @@ install_emscripten() {
     popd || return
 }
 
+install_node
 install_emscripten

.ci/scripts/test_llama.sh

@@ -150,6 +150,7 @@ cmake_install_executorch_libraries() {
     echo "Installing libexecutorch.a, libextension_module.so, libportable_ops_lib.a"
     rm -rf cmake-out
     retry cmake --preset llm \
+        -DBUILD_TESTING=OFF \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
         -DEXECUTORCH_BUILD_QNN="$QNN" \

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -29,6 +29,8 @@ cmake -DPYTHON_EXECUTABLE=python \
     -DEXECUTORCH_ENABLE_LOGGING=1 \
     -DCMAKE_BUILD_TYPE=Release \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
     -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
     -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
     -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \

.ci/scripts/test_llava.sh

@@ -38,6 +38,8 @@ EXECUTORCH_COMMON_CMAKE_ARGS="                      \
         -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON      \
         -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON      \
         -DEXECUTORCH_BUILD_KERNELS_LLM=ON        \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON     \

.ci/scripts/test_model.sh

@@ -188,6 +188,22 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=edsr
     # Additional deps for edsr
     pip install piq
+  elif [[ "${MODEL_NAME}" == "cvt" ]]; then
+    EXPORT_SCRIPT=cvt
+  elif [[ "${MODEL_NAME}" == "dit" ]]; then
+    EXPORT_SCRIPT=dit
+  elif [[ "${MODEL_NAME}" == "efficientnet" ]]; then
+    EXPORT_SCRIPT=efficientnet
+  elif [[ "${MODEL_NAME}" == "focalnet" ]]; then
+    EXPORT_SCRIPT=focalnet
+  elif [[ "${MODEL_NAME}" == "mobilevit_v1" ]]; then
+    EXPORT_SCRIPT=mobilevit_v1
+  elif [[ "${MODEL_NAME}" == "mobilevit_v2" ]]; then
+    EXPORT_SCRIPT=mobilevit_v2
+  elif [[ "${MODEL_NAME}" == "pvt" ]]; then
+    EXPORT_SCRIPT=pvt
+  elif [[ "${MODEL_NAME}" == "swin" ]]; then
+    EXPORT_SCRIPT=swin_transformer
   elif [[ "${MODEL_NAME}" == "albert" ]]; then
     EXPORT_SCRIPT=albert
   elif [[ "${MODEL_NAME}" == "bert" ]]; then
@@ -196,6 +212,8 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=distilbert
   elif [[ "${MODEL_NAME}" == "eurobert" ]]; then
     EXPORT_SCRIPT=eurobert
+  elif [[ "${MODEL_NAME}" == "roberta" ]]; then
+    EXPORT_SCRIPT=roberta
   else
     echo "Unsupported model $MODEL_NAME"
     exit 1
@@ -210,10 +228,13 @@ test_model_with_qnn() {
     "dl3"|"mv3"|"mv2"|"ic4"|"ic3"|"vit"|"mb"|"w2l")
         SCRIPT_FOLDER=scripts
         ;;
-    "albert"|"bert"|"distilbert")
+    "cvt"|"dit"|"focalnet"|"mobilevit_v2"|"pvt"|"swin")
+        SCRIPT_FOLDER=oss_scripts
+        ;;
+    "albert"|"bert"|"distilbert"|"roberta"|"efficientnet"|"mobilevit_v1")
         pip install evaluate
         SCRIPT_FOLDER=oss_scripts
-        # Bert models running in 16bit will encounter op validation fail on some operations,
+        # 16bit models will encounter op validation fail on some operations,
         # which requires CHIPSET >= SM8550.
         QNN_CHIPSET=SM8550
         ;;
@@ -232,21 +253,24 @@ test_model_with_qnn() {
 # @param should_test If true, build and test the model using the coreml_executor_runner.
 test_model_with_coreml() {
   local should_test="$1"
+  local test_with_pybindings="$2"
+  local dtype="$3"
 
   if [[ "${BUILD_TOOL}" != "cmake" ]]; then
     echo "coreml only supports cmake."
     exit 1
   fi
 
-  DTYPE=float16
+  RUN_WITH_PYBINDINGS=""
+  if [[ "${test_with_pybindings}" == true ]]; then
+    echo \"Running with pybindings\"
+    export RUN_WITH_PYBINDINGS="--run_with_pybindings"
+  fi
 
-  "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision "${DTYPE}" --use_partitioner
+  "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision ${dtype} --use_partitioner ${RUN_WITH_PYBINDINGS}
   EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
 
   if [ -n "$EXPORTED_MODEL" ]; then
-    EXPORTED_MODEL_WITH_DTYPE="${EXPORTED_MODEL%.pte}_${DTYPE}.pte"
-    mv "$EXPORTED_MODEL" "$EXPORTED_MODEL_WITH_DTYPE"
-    EXPORTED_MODEL="$EXPORTED_MODEL_WITH_DTYPE"
     echo "OK exported model: $EXPORTED_MODEL"
   else
     echo "[error] failed to export model: no .pte file found"
@@ -303,7 +327,15 @@ elif [[ "${BACKEND}" == *"coreml"* ]]; then
   if [[ "${BACKEND}" == *"test"* ]]; then
     should_test_coreml=true
   fi
-  test_model_with_coreml "${should_test_coreml}"
+  test_with_pybindings=false
+  if [[ "${BACKEND}" == *"pybind"* ]]; then
+    test_with_pybindings=true
+  fi
+  dtype=float16
+  if [[ "${BACKEND}" == *"float32"* ]]; then
+    dtype=float32
+  fi
+  test_model_with_coreml "${should_test_coreml}" "${test_with_pybindings}" "${dtype}"
   if [[ $? -eq 0 ]]; then
     prepare_artifacts_upload
   fi