Update

[ghstack-poisoned]

Author: swolchok

.buckconfig

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

.ci/docker/ci_commit_pins/buck2.txt

@@ -1 +1 @@
-2024-12-16
+2025-05-06

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-7ae0ce6360b6e4f944906502d20da24c04debee5
+01f1cc44cbbfdf6307aa01b803a4ee22f9ade946

.ci/docker/requirements-ci.txt

@@ -17,6 +17,7 @@ parameterized==0.9.0
 
 # Doc build requirements, same as https://github.com/pytorch/pytorch/blob/main/.ci/docker/requirements-docs.txt
 sphinx==5.3.0
+sphinx-reredirects==0.1.4
 sphinx-gallery==0.14.0
 breathe==4.34.0
 exhale==0.2.3

.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() {
@@ -33,13 +28,13 @@ set_up_aot() {
   cmake .. \
       -DCMAKE_INSTALL_PREFIX=$PWD \
       -DEXECUTORCH_BUILD_QNN=ON \
+      -DANDROID_NATIVE_API_LEVEL=30 \
       -DQNN_SDK_ROOT=${QNN_SDK_ROOT} \
       -DEXECUTORCH_BUILD_DEVTOOLS=ON \
       -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/build_android_instrumentation.sh

@@ -22,18 +22,12 @@ install_executorch_and_backend_lib() {
   ANDROID_NDK=/opt/ndk
   BUCK2=buck2
   ANDROID_ABI=arm64-v8a
-  cmake -DBUCK2="${BUCK2}" \
+  cmake --preset llm \
+    -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" \
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
     -DCMAKE_BUILD_TYPE=Release \
-    -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
-    -DEXECUTORCH_BUILD_XNNPACK=ON \
-    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
     -DXNNPACK_ENABLE_ARM_BF16=OFF \
     -Bcmake-android-out .
 
@@ -51,11 +45,7 @@ build_llama_runner() {
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
-    -DCMAKE_BUILD_TYPE=Release -DPYTHON_EXECUTABLE=python \
-    -DEXECUTORCH_BUILD_XNNPACK=ON \
-    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -DCMAKE_BUILD_TYPE=Release \
     -Bcmake-android-out/examples/models/llama examples/models/llama
 
     cmake --build cmake-android-out/examples/models/llama -j4 --config Release

.ci/scripts/build_llama_android.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -eux
+ls pytorch/.git || git clone https://github.com/pytorch/pytorch.git
+pytorch_pin="$(< .ci/docker/ci_commit_pins/pytorch.txt)"
+pushd pytorch
+git checkout "$pytorch_pin"
+popd
+"$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/c10 pytorch/c10

.ci/scripts/check_c10_sync.sh

@@ -0,0 +1,54 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -eux
+
+# Check if dir1's files are also found in dir2 with the same
+# contents. Exempt files named BUCK, CMakeLists.txt, TARGETS, or
+# targets.bzl.
+
+if [ $# -ne 2 ]; then
+    echo "Usage: $0 dir1 dir2" >&2
+    exit 1
+fi
+dir1="$1"
+dir2="$2"
+
+if [ ! -d "$dir1" ] || [ ! -d "$dir2" ]; then
+    echo "Error: Both directories must exist" >&2
+    exit 1
+fi
+
+exit_status=0
+while IFS= read -r -d '' file; do
+    base=$(basename "$file")
+    case "$base" in
+        "BUCK"|"CMakeLists.txt"|"TARGETS"|"targets.bzl")
+            continue
+            ;;
+    esac
+    # Construct the corresponding path in the second directory
+    file2="$dir2/${file#$dir1/}"
+    # Check if the corresponding file exists in the second directory
+    if [ ! -f "$file2" ]; then
+        echo "Error: File '$file' found in '$dir1' but not found in '$dir2'" >&2
+        exit 1
+    fi
+    # Compare the contents of the two files using diff
+    set +ex
+    differences=$(diff -u -p "$file" "$file2")
+    set -e # leave x off
+    # If there are any differences, print an error message and exit with failure status
+    if [ -n "$differences" ]; then
+        echo "Error: Mismatch detected in file '$file':" >&2
+        echo "$differences" >&2
+        exit_status=1
+    fi
+    set -x
+done < <(find "$dir1" -type f -print0)
+
+exit $exit_status

.ci/scripts/compare_dirs.sh

@@ -21,8 +21,11 @@
     "apple_iphone_15": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/3b5acd2e-92e2-4778-b651-7726bafe129d",
     "apple_iphone_15+ios_18": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/12c8b15c-8d03-4e07-950d-0a627e7595b4",
     "samsung_galaxy_s22": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/e59f866a-30aa-4aa1-87b7-4510e5820dfa",
+    "samsung_galaxy_s22_private": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/ea6b049d-1508-4233-9a56-5d9eacbe1078",
     "samsung_galaxy_s24": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98f8788c-2e25-4a3c-8bb2-0d1e8897c0db",
     "google_pixel_8_pro": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/d65096ab-900b-4521-be8b-a3619b69236a",
+    "google_pixel_3_private_rooted": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98d23ca8-ea9e-4fb7-b725-d402017b198d",
+    "apple_iphone_15_private": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/55929353-2f28-4ee5-bdff-d1a95f58cb28",
 }
 
 # Predefined benchmark configurations

.ci/scripts/gather_benchmark_configs.py

@@ -14,7 +14,7 @@
 from typing import Any
 
 from examples.models import MODEL_NAME_TO_MODEL
-from examples.xnnpack import MODEL_NAME_TO_OPTIONS
+from examples.xnnpack import MODEL_NAME_TO_OPTIONS, QuantType
 
 DEFAULT_RUNNERS = {
     "linux": "linux.2xlarge",
@@ -154,7 +154,7 @@ def export_models_for_ci() -> dict[str, dict]:
         if backend == "xnnpack":
             if name not in MODEL_NAME_TO_OPTIONS:
                 continue
-            if MODEL_NAME_TO_OPTIONS[name].quantization:
+            if MODEL_NAME_TO_OPTIONS[name].quantization != QuantType.NONE:
                 backend += "-quantization"
 
             if MODEL_NAME_TO_OPTIONS[name].delegation:

.ci/scripts/gather_test_models.py

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+git clone https://github.com/openvinotoolkit/openvino.git
+cd openvino && git checkout releases/2025/1
+git submodule update --init --recursive
+sudo ./install_build_dependencies.sh
+mkdir build && cd build
+cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_PYTHON=ON
+make -j$(nproc)
+
+cd ..
+cmake --install build --prefix dist
+
+source dist/setupvars.sh
+cd ../backends/openvino
+pip install -r requirements.txt
+cd scripts
+./openvino_build.sh --enable_python

.ci/scripts/setup-openvino.sh

@@ -7,7 +7,7 @@
 
 set -e
 
-APP_PATH="examples/demo-apps/apple_ios/ExecuTorchDemo/ExecuTorchDemo"
+APP_PATH="executorch-examples/mv3/apple/ExecuTorchDemo/ExecuTorchDemo"
 MODEL_NAME="mv3"
 SIMULATOR_NAME="executorch"
 
@@ -34,6 +34,10 @@ say() {
   echo -e "\033[1m\n\t** $1 **\n\033[0m"
 }
 
+say "Cloning the Demo App"
+
+git clone --depth 1 https://github.com/pytorch-labs/executorch-examples.git
+
 say "Installing CoreML Backend Requirements"
 
 ./backends/apple/coreml/scripts/install_requirements.sh

.ci/scripts/test_ios_ci.sh

@@ -152,21 +152,11 @@ which "${PYTHON_EXECUTABLE}"
 cmake_install_executorch_libraries() {
     echo "Installing libexecutorch.a, libextension_module.so, libportable_ops_lib.a"
     rm -rf cmake-out
-    retry cmake \
+    retry cmake --preset llm \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
-        -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-        -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
-        -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM="$CUSTOM" \
-        -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-        -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-        -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
-        -DEXECUTORCH_BUILD_MPS="$MPS" \
-        -DEXECUTORCH_BUILD_COREML="$COREML" \
         -DEXECUTORCH_BUILD_QNN="$QNN" \
         -DQNN_SDK_ROOT="$QNN_SDK_ROOT" \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -Bcmake-out .
     cmake --build cmake-out -j9 --target install --config "$CMAKE_BUILD_TYPE"
 }
@@ -181,10 +171,6 @@ cmake_build_llama_runner() {
     retry cmake \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM="$CUSTOM" \
-        -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-        -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -Bcmake-out/${dir} \
         ${dir}
     cmake --build cmake-out/${dir} -j9 --config "$CMAKE_BUILD_TYPE"
@@ -269,7 +255,7 @@ $PYTHON_EXECUTABLE -m examples.models.llama.export_llama ${EXPORT_ARGS}
 
 # Create tokenizer.bin.
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 
 RUNTIME_ARGS="--model_path=${EXPORTED_MODEL_NAME} --tokenizer_path=tokenizer.bin --prompt=Once --temperature=0 --seq_len=10 --warmup=1"

.ci/scripts/test_llama.sh

@@ -55,7 +55,7 @@ cmake --build cmake-out/examples/models/llama -j16 --config Release
 download_stories_model_artifacts
 
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 # Export model
 LLAMA_CHECKPOINT=stories110M.pt

.ci/scripts/test_llama_torchao_lowbit.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:"
-    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