Merge branch 'main' into change-1091845

Author: zingo

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

@@ -1 +1 @@
-bd06b54e627fbfd354a2cffa4c80fb21883209a9
+44d8d54e38c0258357d4e92e1fefe21e845947a3

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

@@ -38,6 +38,7 @@ set_up_aot() {
       -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM=ON \
       -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM_RUNNER=ON \
       -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
       -DPYTHON_EXECUTABLE=python3

.ci/scripts/setup-samsung-linux-deps.sh

@@ -13,7 +13,7 @@ download_ai_lite_core() {
   API_BASE="https://soc-developer.semiconductor.samsung.com/api/v1/resource/ai-litecore/download"
   API_KEY=$SAMSUNG_AI_LITECORE_KEY
 
-  VERSION="0.5"
+  VERSION="0.7"
   OS_NAME="Ubuntu 22.04"
   OUT_FILE="/tmp/exynos-ai-litecore-v${VERSION}.tar.gz"
   TARGET_PATH="/tmp/exynos_ai_lite_core"
@@ -62,7 +62,7 @@ install_enn_backend() {
   export PYTHONPATH=${PYTHONPATH:-}:${EXECUTORCH_ROOT}/..
 }
 
-AI_LITE_CORE_VERSION=0.5.0
+AI_LITE_CORE_VERSION=0.7.0
 
 download_ai_lite_core ${AI_LITE_CORE_VERSION}
 install_enn_backend

.ci/scripts/test_backend.sh

@@ -59,6 +59,7 @@ fi
 if [[ "$FLOW" == *arm* ]]; then
     # Setup ARM deps.
     .ci/scripts/setup-arm-baremetal-tools.sh
+    source examples/arm/ethos-u-scratch/setup_path.sh
 
     if [[ "$FLOW" == *ethos_u* ]]; then
         # Prepare a test runner binary that can run on the Corstone-3x0 FVPs

.ci/scripts/test_ios_ci.sh

@@ -36,6 +36,7 @@ say() {
 
 say "Cloning the Demo App"
 
+git config --global http.postBuffer 524288000
 git clone --depth 1 https://github.com/meta-pytorch/executorch-examples.git
 
 say "Installing CoreML Backend Requirements"

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -31,6 +31,7 @@ cmake -DPYTHON_EXECUTABLE=python \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
     -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
     -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP=ON \
     -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
     -DEXECUTORCH_BUILD_XNNPACK=OFF \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \

.ci/scripts/test_llava.sh

@@ -38,6 +38,7 @@ EXECUTORCH_COMMON_CMAKE_ARGS="                      \
         -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON      \
         -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP=ON \
         -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
         -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON      \

.ci/scripts/test_model.sh

@@ -48,22 +48,33 @@ prepare_artifacts_upload() {
   fi
 }
 
+
 build_cmake_executor_runner() {
   local backend_string_select="${1:-}"
   echo "Building executor_runner"
   rm -rf ${CMAKE_OUTPUT_DIR}
   mkdir ${CMAKE_OUTPUT_DIR}
+  # Common options:
+  COMMON="-DPYTHON_EXECUTABLE=$PYTHON_EXECUTABLE"
   if [[ "$backend_string_select" == "XNNPACK" ]]; then
     echo "Backend $backend_string_select selected"
-    (cd ${CMAKE_OUTPUT_DIR} \
-      && cmake -DCMAKE_BUILD_TYPE=Release \
+    cmake -DCMAKE_BUILD_TYPE=Release \
         -DEXECUTORCH_BUILD_XNNPACK=ON \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" ..)
+        ${COMMON} \
+        -B${CMAKE_OUTPUT_DIR} .
+    cmake --build ${CMAKE_OUTPUT_DIR} -j4
+  elif [[ "$backend_string_select" == "CUDA" ]]; then
+    echo "Backend $backend_string_select selected"
+    cmake -DCMAKE_BUILD_TYPE=Release \
+        -DEXECUTORCH_BUILD_CUDA=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
+        ${COMMON} \
+        -B${CMAKE_OUTPUT_DIR} .
     cmake --build ${CMAKE_OUTPUT_DIR} -j4
   else
     cmake -DCMAKE_BUILD_TYPE=Debug \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
+        ${COMMON} \
         -B${CMAKE_OUTPUT_DIR} .
     cmake --build ${CMAKE_OUTPUT_DIR} -j4 --config Debug
   fi
@@ -320,6 +331,13 @@ test_model_with_mediatek() {
   EXPORTED_MODEL=$(find "./${EXPORT_SCRIPT}" -type f -name "*.pte" -print -quit)
 }
 
+test_model_with_cuda() {
+  # Export a basic .pte and .ptd, then run the model.
+  "${PYTHON_EXECUTABLE}" -m examples.cuda.scripts.export --model_name="${MODEL_NAME}" --output_dir "./"
+  build_cmake_executor_runner "CUDA"
+  ./${CMAKE_OUTPUT_DIR}/executor_runner --model_path "./${MODEL_NAME}.pte" --data_path "./aoti_cuda_blob.ptd"
+}
+
 
 if [[ "${BACKEND}" == "portable" ]]; then
   echo "Testing ${MODEL_NAME} with portable kernels..."
@@ -372,6 +390,12 @@ elif [[ "${BACKEND}" == "mediatek" ]]; then
   if [[ $? -eq 0 ]]; then
     prepare_artifacts_upload
   fi
+elif [[ "${BACKEND}" == "cuda" ]]; then
+  echo "Testing ${MODEL_NAME} with cuda..."
+  test_model_with_cuda
+  if [[ $? -eq 0 ]]; then
+    prepare_artifacts_upload
+  fi
 else
   set +e
   if [[ "${BACKEND}" == *"quantization"* ]]; then

.ci/scripts/test_torchao_huggingface_checkpoints.sh

@@ -129,6 +129,7 @@ if [[ "$TEST_WITH_RUNNER" -eq 1 ]]; then
         -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
         -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP=ON \
         -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
         -DEXECUTORCH_BUILD_XNNPACK=ON \
         -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \

.ci/scripts/test_wheel_package_qnn.sh

@@ -158,10 +158,12 @@ PY
 
   # Install torchao based on the pinned PyTorch version
   "$PIPBIN" install torch=="${TORCH_VERSION}.${NIGHTLY_VERSION}" --index-url "https://download.pytorch.org/whl/nightly/cpu"
+  "$PIPBIN" install wheel
 
   # Install torchao based on the pinned commit from third-party/ao submodule
   pushd "$REPO_ROOT/third-party/ao" > /dev/null
-  USE_CPP=0 "$PYBIN" setup.py develop
+  export USE_CPP=0
+  "$PIPBIN" install . --no-build-isolation
   popd > /dev/null
 
   echo "=== [$LABEL] Import smoke tests ==="