Merge branch 'main' into check-tensor-dim-in-IR-verifier

Author: digantdesai

.buckconfig

@@ -39,6 +39,7 @@
 
 [buck2]
 restarter=true
+file_watcher=notify
 
 [oss]
 folly_cxx_tests = False

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

@@ -11,17 +11,12 @@ set -o xtrace
 
 build_qnn_backend() {
   echo "Start building qnn backend."
-  export ANDROID_NDK_ROOT=/opt/ndk
-  export QNN_SDK_ROOT=/tmp/qnn/2.28.0.241029
+  export ANDROID_NDK_ROOT=${ANDROID_NDK_ROOT:-/opt/ndk}
+  export QNN_SDK_ROOT=${QNN_SDK_ROOT:-/tmp/qnn/2.28.0.241029}
   export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
 
-  # Workaround to avoid issues around missing flatccrt library (depending on the
-  # number of jobs used), see issue #7300:
-  # Build twice (second time with `--no_clean`) to make sure libflatccrt.a is
-  # available.
-  # TODO: Remove this workaround once the underlying issue is fixed.
-  bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number 2 --release || \
-  bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number 2 --release --no_clean
+  parallelism=$(( $(nproc) - 1 ))
+  bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number ${parallelism} --release
 }
 
 set_up_aot() {
@@ -39,8 +34,7 @@ set_up_aot() {
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
-      -DPYTHON_EXECUTABLE=python3 \
-      -DEXECUTORCH_SEPARATE_FLATCC_HOST_PROJECT=OFF
+      -DPYTHON_EXECUTABLE=python3
   cmake --build $PWD --target "PyQnnManagerAdaptor" "PyQnnWrapperAdaptor" -j$(nproc)
   # install Python APIs to correct import path
   # The filename might vary depending on your Python and host version.

.ci/scripts/test_llava.sh

@@ -93,8 +93,7 @@ cmake_build_llava_runner_for_android() {
         -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK/build/cmake/android.toolchain.cmake \
         -DANDROID_ABI=arm64-v8a                                                 \
         ${LLAVA_COMMON_CMAKE_ARGS}                                              \
-        -DCMAKE_PREFIX_PATH="$python_lib"                  \
-        -DLLAVA_RUNNER_NO_TORCH_DUMMY_IMAGE=ON                                  \
+        -DCMAKE_PREFIX_PATH="$python_lib"                                       \
         -B${BUILD_DIR}/${dir}                                                   \
         ${dir}
 
@@ -107,11 +106,10 @@ export_llava() {
     $PYTHON_EXECUTABLE -m executorch.examples.models.llava.export_llava --pte-name llava.pte --with-artifacts
 }
 
-# Download a new image with different size, to test if the model can handle different image sizes
-prepare_image_tensor() {
+# Download a new image
+download_image() {
     echo "Downloading image"
     curl -o basketball.jpg https://upload.wikimedia.org/wikipedia/commons/7/73/Chicago_Bulls_and_New_Jersey_Nets%2C_March_28%2C_1991.jpg
-    $PYTHON_EXECUTABLE -m executorch.examples.models.llava.image_util --image-path basketball.jpg --output-path image.pt
 }
 
 run_and_verify() {
@@ -121,20 +119,18 @@ run_and_verify() {
         echo "Export failed. Abort"
         exit 1
     fi
-    if [[ ! -f "image.pt" ]]; then
-        echo "image.pt is missing."
+    if [[ ! -f "basketball.jpg" ]]; then
+        echo "basketball.jpg is missing."
         exit 1
     fi
     if [[ ! -f "tokenizer.bin" ]]; then
         echo "tokenizer.bin is missing."
         exit 1
     fi
 
-
-
     RUNTIME_ARGS="--model_path=llava.pte    \
         --tokenizer_path=tokenizer.bin      \
-        --image_path=image.pt               \
+        --image_path=basketball.jpg         \
         --prompt=ASSISTANT:                 \
         --temperature=0                     \
         --seq_len=650"
@@ -149,13 +145,8 @@ run_and_verify() {
 
     # verify result.txt
     RESULT=$(cat result.txt)
-    # set the expected prefix to be the same as prompt because there's a bug in sdpa_with_kv_cache that causes <unk> tokens.
-    if [[ "$(uname)" == "Darwin" ]]; then
-        EXPECTED_PREFIX="ASSISTANT: image captures a basketball game in progress, with several players on the court. One of the players is dribbling the ball, while the others are in various"
-    else
-        # set the expected prefix to be the same as prompt because there's a bug in sdpa_with_kv_cache that causes <unk> tokens.
-        EXPECTED_PREFIX="ASSISTANT: image"
-    fi
+    EXPECTED_PREFIX="ASSISTANT: image captures a basketball game in progress, with several players on the court. "
+
     if [[ "${RESULT}" == *"${EXPECTED_PREFIX}"* ]]; then
         echo "Expected result prefix: ${EXPECTED_PREFIX}"
         echo "Actual result: ${RESULT}"
@@ -184,5 +175,5 @@ fi
 export_llava
 
 # Step3. Run
-prepare_image_tensor
+download_image
 run_and_verify

.ci/scripts/test_model.sh

@@ -87,10 +87,6 @@ test_model() {
     bash examples/models/llava/install_requirements.sh
     STRICT="--no-strict"
   fi
-  if [[ "$MODEL_NAME" == "llama3_2_vision_encoder" || "$MODEL_NAME" == "llama3_2_text_decoder" ]]; then
-    # Install requirements for llama vision.
-    bash examples/models/llama3_2_vision/install_requirements.sh
-  fi
   if [[ "${MODEL_NAME}" == "qwen2_5" ]]; then
       # Install requirements for export_llama
       bash examples/models/llama/install_requirements.sh
@@ -205,7 +201,7 @@ test_model_with_qnn() {
   # TODO(guangyang): Make QNN chipset matches the target device
   QNN_CHIPSET=SM8450
 
-  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --compile_only $EXTRA_FLAGS
+  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --ci --compile_only $EXTRA_FLAGS
   EXPORTED_MODEL=$(find "./${EXPORT_SCRIPT}" -type f -name "${MODEL_NAME}*.pte" -print -quit)
 }
 

.ci/scripts/tests/test_gather_benchmark_configs.py

@@ -53,11 +53,11 @@ def test_skip_disabled_configs(self):
                 "mv3": [
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config1",
-                        github_issue="https://github.com/org/repo/issues/123",
+                        github_issue="https://github.com/org/repo/issues/123",  # @lint-ignore
                     ),
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config2",
-                        github_issue="https://github.com/org/repo/issues/124",
+                        github_issue="https://github.com/org/repo/issues/124",  # @lint-ignore
                     ),
                 ]
             },
@@ -84,7 +84,9 @@ def test_skip_disabled_configs(self):
             self.assertIn("enabled_config2", result)
 
     def test_disabled_configs_have_github_links(self):
-        github_issue_regex = re.compile(r"https://github\.com/.+/.+/issues/\d+")
+        github_issue_regex = re.compile(
+            r"https://github\.com/.+/.+/issues/\d+"  # @lint-ignore
+        )
 
         for (
             model_name,

.ci/scripts/unittest-linux.sh

@@ -20,13 +20,11 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Setup swiftshader and Vulkan SDK which are required to build the Vulkan delegate
     source .ci/scripts/setup-vulkan-linux-deps.sh
 
+    # We need the runner to test the built library.
     PYTHON_EXECUTABLE=python \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
-    # Install llama3_2_vision dependencies.
-    PYTHON_EXECUTABLE=python ./examples/models/llama3_2_vision/install_requirements.sh
-
     .ci/scripts/unittest-linux-cmake.sh
 elif [[ "$BUILD_TOOL" == "buck2" ]]; then
     # Removing this breaks sccache in the Buck build, apparently

.ci/scripts/unittest-macos.sh

@@ -20,16 +20,16 @@ export PATH="${TMP_DIR}:$PATH"
 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_PYBIND=ON -DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"
 
 if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Install llama3_2_vision dependencies.
     PYTHON_EXECUTABLE=python \
     ${CONDA_RUN} --no-capture-output \
-    ./examples/models/llama3_2_vision/install_requirements.sh
 
     .ci/scripts/unittest-macos-cmake.sh
 elif [[ "$BUILD_TOOL" == "buck2" ]]; then

.ci/scripts/wheel/envvar_base.sh

@@ -11,7 +11,3 @@
 # Ensure that CMAKE_ARGS is defined before referencing it. Defaults to empty
 # if not defined.
 export CMAKE_ARGS="${CMAKE_ARGS:-}"
-
-# Link the XNNPACK backend into the pybindings runtime so that users can execute
-# ExecuTorch programs that delegate to it.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON"

.ci/scripts/wheel/envvar_macos.sh

@@ -9,7 +9,3 @@
 # any variables so that subprocesses will see them.
 
 source "${GITHUB_WORKSPACE}/${REPOSITORY}/.ci/scripts/wheel/envvar_base.sh"
-
-# When building for macOS, link additional backends into the pybindings runtime.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_COREML=ON"
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_MPS=ON"

.github/scripts/label_utils.py

@@ -22,12 +22,12 @@
 
 LABEL_ERR_MSG_TITLE = "This PR needs a `release notes:` label"
 LABEL_ERR_MSG = f"""# {LABEL_ERR_MSG_TITLE}
-If your changes are user facing and intended to be a part of release notes, please use a label starting with `release notes:`.
+If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with `release notes:`.
 
-If not, please add the `topic: not user facing` label.
+If not, please add the `release notes: none` label.
 
 To add a label, you can comment to pytorchbot, for example
-`@pytorchbot label "topic: not user facing"`
+`@pytorchbot label "release notes: none"`
 
 For more information, see
 https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.
@@ -51,7 +51,6 @@ def get_last_page_num_from_header(header: Any) -> int:
     # rel="next", <https://api.github.com/repositories/65600975/labels?per_page=100&page=3>; rel="last"
     link_info = header["link"]
     # Docs does not specify that it should be present for projects with just few labels
-    # And https://github.com/malfet/deleteme/actions/runs/7334565243/job/19971396887 it's not the case
     if link_info is None:
         return 1
     prefix = "&page="
@@ -115,7 +114,7 @@ def has_required_labels(pr: "GitHubPR") -> bool:
     pr_labels = pr.get_labels()
     # Check if PR is not user facing
     is_not_user_facing_pr = any(
-        label.strip() == "topic: not user facing" for label in pr_labels
+        label.strip() == "release notes: none" for label in pr_labels
     )
     return is_not_user_facing_pr or any(
         label.strip() in get_release_notes_labels(pr.org, pr.project)