Update

[ghstack-poisoned]

Author: mcr229

.ci/docker/ci_commit_pins/pytorch.txt

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

.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-qnn-deps.sh

@@ -7,47 +7,7 @@
 
 set -ex
 
-verify_pkg_installed() {
-  echo $(dpkg-query -W --showformat='${Status}\n' $1|grep "install ok installed")
-}
+source "$(dirname "${BASH_SOURCE[0]}")/../../backends/qualcomm/scripts/install_qnn_sdk.sh"
 
-install_qnn() {
-  echo "Start installing qnn."
-  QNN_INSTALLATION_DIR=/tmp/qnn
-  mkdir -p "${QNN_INSTALLATION_DIR}"
-
-  curl -Lo /tmp/v2.28.0.24.10.29.zip "https://softwarecenter.qualcomm.com/api/download/software/qualcomm_neural_processing_sdk/v2.28.0.241029.zip"
-  echo "Finishing downloading qnn sdk."
-  unzip -qo /tmp/v2.28.0.24.10.29.zip -d /tmp
-  echo "Finishing unzip qnn sdk."
-
-
-  # Print the content for manual verification
-  ls -lah "/tmp/qairt"
-  mv "/tmp/qairt"/* "${QNN_INSTALLATION_DIR}"
-  echo "Finishing installing qnn '${QNN_INSTALLATION_DIR}' ."
-
-  ls -lah "${QNN_INSTALLATION_DIR}"
-}
-
-setup_libc++() {
-  clang_version=$1
-  sudo apt-get update
-  pkgs_to_check=("libc++-${clang_version}-dev")
-  j=0
-  while [ $j -lt ${#pkgs_to_check[*]} ]; do
-    install_status=$(verify_pkg_installed ${pkgs_to_check[$j]})
-    if [ "$install_status" == "" ]; then
-      sudo apt-get install -y ${pkgs_to_check[$j]}
-      if [[ $? -ne 0 ]]; then
-        echo "ERROR: Failed to install required packages for libc++"
-        exit 1
-      fi
-    fi
-    j=$(( $j +1));
-  done
-}
-
-# This needs to match with the clang version from the Docker image
-setup_libc++ 12
+setup_libcpp 12
 install_qnn

.ci/scripts/test_ane_static_llama.sh

@@ -28,6 +28,6 @@ pushd $EXECUTORCH_ROOT/examples/apple/coreml/llama
 # Download stories llama110m artifacts
 download_stories_model_artifacts
 
-python export.py -n model.pte -p params.json -c stories110M.pt --seq_length 32 --max_seq_length 64 --dtype fp16 --coreml-quantize c4w
+python export.py -n model.pte -p params.json -c stories110M.pt --seq_length 32 --max_seq_length 64 --dtype fp16 --coreml-quantize c4w --embedding-quantize 4,32
 
 popd

.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_qnn_static_llama.sh

@@ -41,6 +41,10 @@ exit_code1=$?
 $PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleLLMScript.test_llama_stories_110m --model SM8650 --build_folder build-x86/ --executorch_root . --artifact_dir . --llama_artifacts . --enable_x86_64
 exit_code2=$?
 
+# Check BC
+bash backends/qualcomm/bc/test_qnn_static_llama_bc.sh
+exit_code3=$?
+
 # Check the exit codes and print messages
 if [ $exit_code1 -ne 0 ]; then
     echo "Static Llama compile only with weight sharing test failed. $exit_code1."
@@ -50,8 +54,12 @@ if [ $exit_code2 -ne 0 ]; then
     echo "Static Llama accuracy test failed. $exit_code2."
 fi
 
+if [ $exit_code3 -ne 0 ]; then
+    echo "Static Llama BACKWARD COMPATIBILITY test failed. $exit_code3."
+fi
+
 # Return failure if either program failed
-if [ $exit_code1 -ne 0 ] || [ $exit_code2 -ne 0 ]; then
+if [ $exit_code1 -ne 0 ] || [ $exit_code2 -ne 0 ] || [ $exit_code3 -ne 0 ]; then
     exit 1
 else
     exit 0

.ci/scripts/unittest-linux.sh

@@ -22,7 +22,7 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
 
     # We need the runner to test the built library.
     PYTHON_EXECUTABLE=python \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL=ON -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
     .ci/scripts/unittest-linux-cmake.sh

.ci/scripts/unittest-macos.sh

@@ -22,7 +22,7 @@ trap 'rm -rfv ${TMP_DIR}' EXIT
 # Setup MacOS dependencies as there is no Docker support on MacOS atm
 # We need the runner to test the built library.
 PYTHON_EXECUTABLE=python \
-CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL=ON -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"