Merge branch 'main' into move_ethosu

Author: digantdesai

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

@@ -0,0 +1 @@
+a3942627f5ac048e06b4b1d703b0a6a53bf6da5b

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-5616fa4a68718ead203314a3467f7dd9547153ae
+7cda4017ddda554752e89069ae205be5e8388f59

.ci/docker/common/install_conda.sh

@@ -13,7 +13,7 @@ source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 install_miniconda() {
   BASE_URL="https://repo.anaconda.com/miniconda"
   CONDA_FILE="Miniconda3-py${PYTHON_VERSION//./}_${MINICONDA_VERSION}-Linux-x86_64.sh"
-  if [[ $(uname -m) == "aarch64" ]]; then 
+  if [[ $(uname -m) == "aarch64" ]]; then
     CONDA_FILE="Miniconda3-py${PYTHON_VERSION//./}_${MINICONDA_VERSION}-Linux-aarch64.sh"
   fi
 
@@ -71,4 +71,8 @@ fix_conda_ubuntu_libstdcxx() {
 install_miniconda
 install_python
 install_pip_dependencies
-fix_conda_ubuntu_libstdcxx
+# Hack breaks the job on aarch64 but is still necessary everywhere
+# else.
+if [ "$(uname -m)" != "aarch64" ]; then
+    fix_conda_ubuntu_libstdcxx
+fi

.ci/docker/conda-env-ci.txt

@@ -1,4 +1,4 @@
-cmake=3.26.4
+cmake=3.31.2
 ninja=1.10.2
 libuv
 llvm-openmp

.ci/scripts/check_c10_sync.sh

@@ -12,3 +12,4 @@ pushd pytorch
 git checkout "$pytorch_pin"
 popd
 "$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/c10 pytorch/c10
+"$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/torch/headeronly pytorch/torch/headeronly

.ci/scripts/setup-arm-baremetal-tools.sh

@@ -8,4 +8,4 @@
 # Setup arm example environment (including TOSA tools)
 git config --global user.email "[email protected]"
 git config --global user.name "Github Executorch"
-bash examples/arm/setup.sh --i-agree-to-the-contained-eula
+bash examples/arm/setup.sh --i-agree-to-the-contained-eula ${@:-}

.ci/scripts/test_yolo12.sh

@@ -0,0 +1,197 @@
+#!/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"
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    -model)
+      MODEL_NAME="$2" # stories110M
+      shift 2
+      ;;
+    -mode)
+      MODE="$2" # portable or xnnpack+custom or xnnpack+custom+qe
+      shift 2
+      ;;
+    -pt2e_quantize)
+      PT2E_QUANTIZE="$2"
+      shift 2
+      ;;
+    -upload)
+      UPLOAD_DIR="$2"
+      shift 2
+      ;;
+    -video_path)
+      VIDEO_PATH="$2" # portable or xnnpack+custom or xnnpack+custom+qe
+      shift 2
+      ;;
+    *)
+      echo "Unknown option: $1"
+      usage
+      ;;
+  esac
+done
+
+# Default mode to xnnpack+custom if not set
+MODE=${MODE:-"openvino"}
+
+# Default UPLOAD_DIR to empty string if not set
+UPLOAD_DIR="${UPLOAD_DIR:-}"
+
+# Default PT2E_QUANTIZE to empty string if not set
+PT2E_QUANTIZE="${PT2E_QUANTIZE:-}"
+
+# Default CMake Build Type to release mode
+CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-Release}
+
+if [[ $# -lt 5 ]]; then # Assuming 4 mandatory args
+    echo "Expecting atleast 5 positional arguments"
+    echo "Usage: [...]"
+fi
+if [[ -z "${MODEL_NAME:-}" ]]; then
+  echo "Missing model name, exiting..."
+  exit 1
+fi
+
+
+if [[ -z "${MODE:-}" ]]; then
+  echo "Missing mode, choose openvino or xnnpack, exiting..."
+  exit 1
+fi
+
+if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
+  PYTHON_EXECUTABLE=python3
+fi
+
+TARGET_LIBS=""
+
+if [[ "${MODE}" =~ .*openvino.* ]]; then
+  OPENVINO=ON
+  TARGET_LIBS="$TARGET_LIBS openvino_backend "
+
+  git clone https://github.com/openvinotoolkit/openvino.git
+  cd openvino && git b16b776ac119dafda51f69a80f1e6b7376d02c3b
+  git submodule update --init --recursive
+  sudo ./install_build_dependencies.sh
+  mkdir build && cd build
+  cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_PYTHON=ON
+  make -j$(nproc)
+
+  cd ..
+  cmake --install build --prefix dist
+
+  source dist/setupvars.sh
+  cd ../backends/openvino
+  pip install -r requirements.txt
+  cd ../../
+else
+  OPENVINO=OFF
+fi
+
+if [[ "${MODE}" =~ .*xnnpack.* ]]; then
+  XNNPACK=ON
+  TARGET_LIBS="$TARGET_LIBS xnnpack_backend "
+else
+  XNNPACK=OFF
+fi
+
+which "${PYTHON_EXECUTABLE}"
+
+
+DIR="examples/models/yolo12"
+$PYTHON_EXECUTABLE -m pip install -r ${DIR}/requirements.txt
+
+cmake_install_executorch_libraries() {
+    rm -rf cmake-out
+    build_dir=cmake-out
+    mkdir $build_dir
+
+
+    retry cmake -DCMAKE_INSTALL_PREFIX="${build_dir}" \
+          -DCMAKE_BUILD_TYPE="${CMAKE_BUILD_TYPE}" \
+          -DEXECUTORCH_BUILD_OPENVINO="$OPENVINO" \
+          -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
+          -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+          -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+          -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
+          -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
+          -B"${build_dir}"
+
+    # Build the project
+    cmake --build ${build_dir} --target install --config ${CMAKE_BUILD_TYPE} -j$(nproc)
+
+    export CMAKE_ARGS="
+                       -DEXECUTORCH_BUILD_OPENVINO="$OPENVINO" \
+                       -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
+                       -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+                       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+                       -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
+                       -DEXECUTORCH_ENABLE_LOGGING=ON \
+                       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
+                       -DEXECUTORCH_BUILD_PYBIND=ON"
+
+    echo $TARGET_LIBS
+    export CMAKE_BUILD_ARGS="--target $TARGET_LIBS"
+    pip install . --no-build-isolation
+}
+
+cmake_build_demo() {
+    echo "Building yolo12 runner"
+    retry cmake \
+        -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
+        -DUSE_OPENVINO_BACKEND="$OPENVINO" \
+        -DUSE_XNNPACK_BACKEND="$XNNPACK" \
+        -Bcmake-out/${DIR} \
+        ${DIR}
+    cmake --build cmake-out/${DIR} -j9 --config "$CMAKE_BUILD_TYPE"
+
+}
+
+cleanup_files() {
+ rm $EXPORTED_MODEL_NAME
+}
+
+prepare_artifacts_upload() {
+  if [ -n "${UPLOAD_DIR}" ]; then
+    echo "Preparing for uploading generated artifacs"
+    zip -j model.zip "${EXPORTED_MODEL_NAME}"
+    mkdir -p "${UPLOAD_DIR}"
+    mv model.zip "${UPLOAD_DIR}"
+    mv result.txt "${UPLOAD_DIR}"
+
+  fi
+}
+
+
+# Export model.
+EXPORTED_MODEL_NAME="${MODEL_NAME}_fp32_${MODE}.pte"
+echo "Exporting ${EXPORTED_MODEL_NAME}"
+EXPORT_ARGS="--model_name=${MODEL_NAME} --backend=${MODE}"
+
+# Add dynamically linked library location
+cmake_install_executorch_libraries
+
+$PYTHON_EXECUTABLE -m examples.models.yolo12.export_and_validate ${EXPORT_ARGS}
+
+
+RUNTIME_ARGS="--model_path=${EXPORTED_MODEL_NAME} --input_path=${VIDEO_PATH}"
+# Check build tool.
+cmake_build_demo
+# Run yolo12 runner
+NOW=$(date +"%H:%M:%S")
+echo "Starting to run yolo12 runner at ${NOW}"
+# shellcheck source=/dev/null
+cmake-out/examples/models/yolo12/Yolo12DetectionDemo ${RUNTIME_ARGS} > result.txt
+NOW=$(date +"%H:%M:%S")
+echo "Finished at ${NOW}"
+
+RESULT=$(cat result.txt)
+
+prepare_artifacts_upload
+cleanup_files

.github/workflows/android-perf-private-device-experiment.yml

@@ -57,6 +57,6 @@ jobs:
       id-token: write
       contents: read
     with:
-      models: ${{ inputs.models || github.event_name == 'schedule' && 'Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,meta-llama/Llama-3.2-1B,allenai/OLMo-1B-hf' || 'Qwen/Qwen3-0.6B' }}
+      models: ${{ inputs.models || github.event_name == 'schedule' && 'Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,meta-llama/Llama-3.2-1B,allenai/OLMo-1B-hf,google/gemma-3-1b-it' || 'google/gemma-3-1b-it' }}
       devices: samsung_galaxy_s22_private
       benchmark_configs: ${{ inputs.benchmark_configs }}

.github/workflows/android-perf.yml

@@ -72,7 +72,7 @@ jobs:
           # Separate default values from the workflow dispatch. To ensure defaults are accessible
           # during scheduled runs and to provide flexibility for different defaults between
           # on-demand and periodic benchmarking.
-          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l,meta-llama/Llama-3.2-1B,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8,Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,allenai/OLMo-1B-hf' || 'Qwen/Qwen3-0.6B' }}
+          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l,meta-llama/Llama-3.2-1B,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8,Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,allenai/OLMo-1B-hf,google/gemma-3-1b-it' || 'Qwen/Qwen3-0.6B' }}
           CRON_DEFAULT_DEVICES: samsung_galaxy_s22
         run: |
           set -eux
@@ -341,10 +341,11 @@ jobs:
               echo "tokenizer.json is downloaded to $DOWNLOADED_PATH"
 
               # Install optimum-executorch
+              OPTIMUM_ET_COMMIT=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
               git clone https://github.com/huggingface/optimum-executorch
               pushd optimum-executorch
               # There is no release yet, for CI stability, always test from the same commit on main
-              git checkout 4c3b18f6cca68c5ccff809131d570062723d7188
+              git checkout $OPTIMUM_ET_COMMIT
               python install_dev.py --skip_override_torch
               pip list
 
@@ -353,21 +354,12 @@ jobs:
                 "--task" "text-generation"
                 "--recipe" "xnnpack"
                 "--use_custom_sdpa"
+                "--use_custom_kv_cache"
                 "--qlinear"
                 "--qembedding"
                 "--output_dir" ".."
               )
 
-              # Add conditional arguments based on model
-              case "${HF_MODEL_REPO}" in
-                *"google/gemma-3-1b-it"*)
-                  echo "--use_custom_kv_cache can not be used for HybridCache"
-                  ;;
-                *)
-                  ARGS+=("--use_custom_kv_cache")
-                  ;;
-              esac
-
               optimum-cli export executorch "${ARGS[@]}"
               popd
 

.github/workflows/apple-perf-private-device-experiment.yml

@@ -57,6 +57,6 @@ jobs:
       id-token: write
       contents: read
     with:
-      models: ${{ inputs.models || github.event_name == 'schedule' && 'Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,meta-llama/Llama-3.2-1B,allenai/OLMo-1B-hf' || 'Qwen/Qwen3-0.6B' }}
+      models: ${{ inputs.models || github.event_name == 'schedule' && 'Qwen/Qwen3-0.6B,HuggingFaceTB/SmolLM2-135M,meta-llama/Llama-3.2-1B,allenai/OLMo-1B-hf,google/gemma-3-1b-it' || 'google/gemma-3-1b-it' }}
       devices: apple_iphone_15_private
       benchmark_configs: ${{ inputs.benchmark_configs }}